U.S. patent application number 13/437221 was filed with the patent office on 2012-07-19 for method of processing of nitride semiconductor wafer, nitride semiconductor wafer, method of producing nitride semiconductor device and nitride semiconductor device.
This patent application is currently assigned to SUMITOMO ELECTRIC INDUSTRIES, LTD.. Invention is credited to Keiji Ishibashi, Naoki Matsumoto, Hidenori Mikami.
Application Number | 20120184108 13/437221 |
Document ID | / |
Family ID | 41015962 |
Filed Date | 2012-07-19 |
United States Patent
Application |
20120184108 |
Kind Code |
A1 |
Ishibashi; Keiji ; et
al. |
July 19, 2012 |
METHOD OF PROCESSING OF NITRIDE SEMICONDUCTOR WAFER, NITRIDE
SEMICONDUCTOR WAFER, METHOD OF PRODUCING NITRIDE SEMICONDUCTOR
DEVICE AND NITRIDE SEMICONDUCTOR DEVICE
Abstract
A nitride semiconductor wafer is planar-processed by grinding a
bottom surface of the wafer, etching the bottom surface by, e.g.,
KOH for removing a bottom process-induced degradation layer,
chamfering by a rubber whetstone bonded with 100 wt %-60 wt %
#3000-#600 diamond granules and 0 wt %-40 wt % oxide granules,
grinding and polishing a top surface of the wafer, etching the top
surface for eliminating a top process-induced degradation layer and
maintaining a 0.5 .mu.m-10 .mu.m thick edge process-induced
degradation layer.
Inventors: |
Ishibashi; Keiji; (Hyogo,
JP) ; Mikami; Hidenori; (Hyogo, JP) ;
Matsumoto; Naoki; (Hyogo, JP) |
Assignee: |
SUMITOMO ELECTRIC INDUSTRIES,
LTD.
|
Family ID: |
41015962 |
Appl. No.: |
13/437221 |
Filed: |
April 2, 2012 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
13287670 |
Nov 2, 2011 |
8183669 |
|
|
13437221 |
|
|
|
|
12940733 |
Nov 5, 2010 |
8101523 |
|
|
13287670 |
|
|
|
|
12812697 |
Jul 13, 2010 |
7872331 |
|
|
PCT/JP2009/053112 |
Feb 16, 2009 |
|
|
|
12940733 |
|
|
|
|
Current U.S.
Class: |
438/706 ;
257/E21.214; 257/E21.237; 438/689 |
Current CPC
Class: |
H01L 2924/12041
20130101; H01L 2924/12041 20130101; C30B 33/00 20130101; H01L 24/32
20130101; H01L 2224/73265 20130101; H01L 21/02021 20130101; H01L
2224/48091 20130101; Y10S 438/959 20130101; H01L 2224/48091
20130101; H01L 2924/12042 20130101; H01L 2924/12042 20130101; Y10S
438/928 20130101; H01L 2224/48247 20130101; C30B 29/403 20130101;
H01L 2924/00 20130101; H01L 2924/00 20130101; H01L 2924/00014
20130101; Y10S 438/977 20130101; H01L 33/0075 20130101 |
Class at
Publication: |
438/706 ;
438/689; 257/E21.214; 257/E21.237 |
International
Class: |
H01L 21/302 20060101
H01L021/302 |
Foreign Application Data
Date |
Code |
Application Number |
Feb 27, 2008 |
JP |
2008-045425 |
Feb 9, 2009 |
JP |
2009-026834 |
Claims
1-12. (canceled)
13. A method of processing a nitride semiconductor wafer comprising
the steps of: flat-processing a bottom surface of the nitride
semiconductor wafer; eliminating a process-induced degradation
layer on the bottom surface by etching; chamfering an edge of the
wafer by a diamond-bonded whetstone; flat-processing a top surface
of the wafer; eliminating a process-induced degradation layer on
the top surface by vapor-phase etching; and maintaining a 0.5-10
.mu.m thick edge process-induced degradation layer remaining on the
edge.
14. A method of processing a nitride semiconductor wafer comprising
the steps of: slicing a nitride semiconductor crystal ingot into
nitride semiconductor wafers; eliminating a process-induced
degradation layer on a bottom surface by etching; chamfering an
edge of the wafer by a diamond-bonded whetstone; flat-processing a
top surface of the wafer; eliminating a process-induced degradation
layer on the top surface by vapor-phase etching; and maintaining a
0.5-10 .mu.m thick edge process-induced degradation layer remaining
on the edge.
15. The method of processing a nitride semiconductor wafer as
claimed in claim 13 or 14, wherein the whetstone of chamfering is a
rubber-bonding or bubbled-resin-bonding whetstone which includes
#3000-#600 diamond granules at 100 wt %-60 wt % and oxide granules
at 0 wt %-40 wt % and is bonded to a base plate by the rubber or
bubbled resin.
16. The method of processing a nitride semiconductor wafer as
claimed in claim 15, wherein the oxide granules are
Fe.sub.2O.sub.3, Cr.sub.2O.sub.3, MnO.sub.2, ZnO, CuO,
CO.sub.3O.sub.4 or Fe.sub.3O.sub.4.
17. A method of processing a nitride semiconductor wafer comprising
the steps of: grinding a bottom surface of a gallium nitride wafer;
inducing a process-induced degradation layer on the bottom surface;
etching away the bottom process-induced degradation layer with any
one of KOH, NaOH and H.sub.3PO.sub.4 solutions; chamfering an edge
of the gallium nitride wafer by a rubber-bonding whetstone;
inducing an edge process-induced degradation layer on the edge;
grinding a top surface of the gallium nitride wafer; inducing a
process-induced degradation layer on the top surface;
rough-polishing the top surface with rough whetting granules;
fine-polishing the top surface with fine whetting granules;
eliminating a part of the edge process-induced degradation layer;
and maintaining a 1-3 .mu.m thick edge process-induced degradation
layer on the edge.
Description
FIELD OF THE INVENTION
[0001] This invention relates to a chamfering method of nitride
semiconductor wafers, and a nitride semiconductor wafer. Group 3
nitride semiconductors, e.g., gallium nitride (GaN), aluminum
nitride (AlN) and so on, have wide bandgaps and promising
applications to light emitting devices and electronic devices.
Neither large GaN nor AlN single crystal exist in nature. Vapor
phase growth methods can produce nitride semiconductor single
crystals by supplying material gases on an undersubstrate made of a
foreign material and synthesizing nitride semiconductor crystal
films on the undersubstrate in vapor phase. A sodium flux method
can produce nitride semiconductor single crystals by solving
nitrogen gas into metal sodium including gallium, making nitrogen
gas react with gallium and synthesizing GaN crystal in liquid
phase. An ammonothermal method can produce nitride semiconductor
single crystals by making NH.sub.3 react with Ga under the
supercritical state and synthesizing GaN crystal in liquid
phase.
[0002] Since no freestanding large GaN or AlN wafer had been
produced at an early stage, light emitting devices were produced by
epitaxially growing GaN, InGaN, AlGaN and other group 3 nitride
films on sapphire wafers as undersubstrates. Then vapor phase
growth methods or liquid phase growth methods enable us to make
wide freestanding GaN substrate crystals. At present, freestanding
GaN wafers, AlN wafers, AlGaN wafers or InGaN wafers with diameters
of 1 inch (about 25 mm) or 2 inch (about 50 mm) can be
produced.
[0003] As-grown crystals which are made by the vapor growth phase
methods or the liquid phase growth methods have rough surfaces and
large fluctuation of thicknesses. Grinding and polishing of
surfaces of as-cut wafers make mirror wafers. Semiconductor wafers
serve a variety of semiconductor devices as substrates on which the
semiconductor devices are fabricated. Thus the surfaces of
semiconductor wafers should be flat, crack-free and bow-free.
[0004] If peripheral edges of wafers were zigzag, the wafers would
be liable to crack, split or break. Thus peripheral sharp edges of
wafers should be eliminated. The process of eliminating sharp edges
is called "beveling" or "chamfering". Chamfering of nitride
semiconductors, e.g., GaN, which are rigid and fragile, requires
deliberate contrivance.
BACKGROUND OF THE INVENTION
[0005] Prevalent silicon (Si) wafers or gallium arsenide (GaAs)
wafers come on the market mainly as circular wafers for the
convenience of handling. Gallium nitride wafers or other nitride
wafers will be also produced mainly as circular wafers. Top
surfaces of semiconductor wafers shall be polished to obtain mirror
wafers. In accordance with purposes, sometimes both top and bottom
surfaces will be mirror-polished or sometimes only top surfaces
will be mirror-polished.
[0006] In addition to the surface grinding and polishing, edges of
wafers will be ground or polished. As-cut wafers have harp edges
which would cause crack, break and fracture of wafers. Fine
fractions would damage or contaminate top surfaces. For preventing
wafers from cracking or breaking, edges of wafers are ground or
ground/polished. In addition to edges, circumferences (circular
outer sides) are sometimes ground/polished. The processing of
grinding of edges (including peripheries and circumferences) is
called "chamfering" or "bevelling". "Chamfering" and "bevelling"
are synonyms.
[0007] Silicon wafers are also chamfered. Nitride wafers, e.g., Si
wafers, GaN wafers are different from prevalent silicon wafers in
rigidity, fragility, solidity and chemical properties. GaN and
other nitride crystals are more rigid, more fragile than Si. It is
impossible to chemically etch III group planes of nitride, e.g.,
Ga-plane of GaN crystals.
[0008] Si-chamfering technology cannot directly be applied to
GaN-chamfering or other nitride chamfering. Chamfering technology
being mature for silicon wafers is useless for GaN wafers and other
nitride wafers.
[0009] Different kinds of semiconductor wafers require different
chamfering techniques. Different optimum chamfering methods
suitable for different kinds of semiconductor wafers should be
sought out by trial and error.
[0010] Patent document (1) Japanese Patent Laying Open No.
2002-356398, "Gallium nitride wafer", alleged to make first
freestanding GaN substrates, which had not existed, by an ELO
(Epitaxial Lateral Overgrowth) method, chamfer the GaN substrates
and form orientation flats (OF) to the GaN wafers. The ELO method
of Patent document (1) employed (111) GaAs wafer undersubstrates.
The ELO deposited an SiO.sub.2 mask with honeycomb distributing
microwindows on the GaAs wafer, grew a GaN thin film in horizontal
directions on the mask by an HVPE (Hydride Vapor Phase Epitaxy),
decreased dislocations, jointed GaN films on the mask and grew GaN
along the c-axis.
[0011] The HVPE produced more than 100 .mu.m thick GaN crystals,
eliminated the GaAs undersubstrate and obtained a freestanding GaN
substrate. An obtained GaN substrate had nearly a square-shape. A
circular wafer was obtained by grinding the square-shaped as-grown
GaN wafer into a circle. For discriminating orientations and
obverse/reverse surfaces, the wafer was provided with an OF
(orientation flat) and an IF (identification flat) to a (1-100)
plane side or a (2-1-10) plane side.
[0012] Patent document (1) Japanese Patent Laying Open No.
2002-356398, "Gallium nitride wafer", disclosed methods of
chamfering a GaN wafer into a slant edge at an oblique angle of 5
degrees to 30 degrees or into a round edge of a radius of 0.1 mm to
0.5 mm. The former case rotates a circular, conical shaped
whetstone, brings the whetstone into outer contact with an edge of
a GaN wafer, and grinds the sharp edge into a slant of an angle of
5 degrees to 30 degrees. The conical whetstone is a resin-bonding
whetstone fixing stationary whetgranules by a resin to the surface
of the base.
[0013] The latter case rotates a spool-shaped whetstone with an
inner half circular section of a radius of 0.1 mm to 0.5 mm, brings
an edge of a GaN wafer into outer contact with the whetstone and
grinds the edge into a circular section.
[0014] It turned out, however, that chamfering by the resin-bonding
whetstone was liable to induce break or crack in GaN wafers.
"Resin-bonding" means that a resin bonds stationary whetgranules to
the surface of a base.
[0015] (2) Japanese Patent Laying Open No. 2005-136167 (Filing No.
2003-370430), "Method of producing a nitride semiconductor wafer
and Nitride semiconductor wafer", pointed out a problem that GaN
wafers which were produced by growing a thick GaN crystal on a
foreign material undersubstrate, removing the undersubstrate and
obtaining a freestanding GaN crystal were suffering from large bow
caused by big differences of thermal expansion and lattice constant
between the undersubstrate material and GaN. The document (2)
explained that the bow height difference between the center and
circumference reached to .+-.40 .mu.m to .+-.100 .mu.m.
[0016] Polishing a wafer makes a process-induced degradation layer
on the surface of the wafer. The Patent document (2) told that the
process-induced degradation layer introduced by polishing has a
function of expanding the polished surface. The Patent Document (2)
said that the bow of nitride wafers could be reduced by the
function. The document (2) said that thinning the process-induced
degradation layer by etching decreases the expansion force. Further
the document (2) declared that combining both polishing and etching
on top and bottom surfaces would decrease the bow of nitride
wafers.
[0017] When a bottom surface (nitride plane; N-plane) is concavely
deformed, bottom surface polishing expands the bottom surface by
inducing a process-induced degradation layer on the bottom. The
bottom surface is distorted convexly. The direction of bow is
reversed. The reverse bow should be suppressed. Thus the
process-induced degradation layer is thinned by etching. Bottom
expansion stress is reduced with thinning of the bottom
process-induced degradation layer. The bow decreases.
[0018] The document (2) said that the top surface (Ga-plane) was
rigid and difficult to polish. Etching of the top surface was also
difficult. The document (2) taught that dry etching by chloride
plasma was applicable to the top surface (Ga-plane). Grinding and
polishing of a GaN wafer caused a 10 .mu.m thick process-induced
degradation layer on a top surface and a 50 .mu.m thick
process-induced degradation layer on a bottom surface. The document
(2) asserted that the bow could be alleviated by adjusting the
thickness of the bottom process-induced degradation layer within a
pertinent range by polishing and etching.
[0019] Embodiment 1 of the document (2) made a 5 .mu.m top-concave
GaN wafer enjoying less bow by producing a 50 .mu.m top-concave GaN
as-cut wafer, grinding the top surface for reversing the bow to 30
.mu.m top-convex, dry-etching the top for decreasing the bow to 20
.mu.m top-concave, grinding the bottom and etching the bottom. The
document (2) declared that a pertinent series of grinding and
etching could finally reduce the bow of GaN wafers within a range
of +30 .mu.m to -20 .mu.m in a scale of reducing to 2 inch (50 mm)
of diameter.
[0020] Plus sign denoted top-convex bow Minus sign denoted
top-concave bow. The document (2) asserted that the bow of GaN
wafers could be suppressed within the scope from 30 .mu.m
top-convex to 20 .mu.m top-concave.
[0021] (3) Japanese Patent Laying Open No. 2004-319951 (Filing No.
2003-275935), "Edge-polished nitride semiconductor wafer,
edge-polished GaN free standing substrate wafer and method of
processing an edge-polished nitride semiconductor wafer", pointed
out that chamfering of a GaN wafer by a resin-bonding diamond
whetstone being in outer contact with an edge of the GaN wafer
induced cracks or breaks of wafers, since GaN is fragile and hard.
The document (3) denied resin-bond whetstone chamfering.
[0022] The document (3) proposed a whettape chamfering method by
preparing a whettape with a tape having stationary whetgranules,
bringing an edge in a circular direction in inner-contact with the
whettape, rotating the wafer and bevelling the edge by the
whettape. The whettape rolls off one spool onto another, in order
to expose a new part of the whettape surface. When the stationary
granules were worn out, the contacting part is renewed by feeding
the edge with a fresh part of the whettape for maintaining the same
condition of whetting. Since the edge was in inner contact with the
whettape, the contact pressure was small and no shock acted on the
edge. The document (3) alleged that the whettape could enhance the
yield far higher than the outer-contacting resin-bonding
whetstone.
[0023] FIG. 11 is a perspective view of the whettape chamfering
first proposed by the document (3) (Japanese Patent Laying Open No.
2004-319951). FIG. 12 is an enlarged sectional view of a contacting
whettape and an edge of a wafer.
[0024] A wafer W is fixed to a rotary disc (not shown in the
figure) by vacuum chucking. An edge E of the wafer W is in inner
contact with the whettape T. A central angle of the contacting part
is 40 degrees to 90 degrees. The document (3) asserted that soft
contact of the edge with the elastic whettape could prevent the
edge from cracking and breaking. [0025] Document (1)=Japanese
Patent Laying Open No. 2002-356398, "Gallium Nitride wafer" (Filing
No. 2001-166904). [0026] Document (2)=Japanese Patent Laying Open
No. 2005-136167, "Method of producing Nitride semiconductor wafers
and Nitride semiconductor wafer. (Filing No. 2003-370430). [0027]
Document (3)=Japanese Patent Laying Open No. 2004-319951, "Edge
polished Nitride semiconductor wafer and method of processing an
edge polished GaN freestanding wafer and a method of chamfering a
nitride semiconductor wafer" (Filing No. 2003-275935).
[0028] Recent development has enabled the vapor phase growth method
and liquid phase growth method to produce nitride semiconductor
freestanding crystals. It is, however, still difficult for the
liquid phase method to make a large nitride crystal. Large
freestanding nitride semiconductor as-cut wafers can be produced by
growing a thick GaN crystal on a wide undersubstrate, cutting the
GaN crystal by a wiresaw and separating the GaN crystal from the
undersubstrate. A series of bottom grinding, chamfering and top
grinding/polishing convert the as-cut nitride wafers into nitride
semiconductor mirror wafers.
[0029] Grinding induces a thick (10 .mu.m-50 .mu.m) process-induced
degradation layer on the part in contact with a whetstone. Bottom
(surface) grinding causes a process-induced degradation layer on
the bottom surface. Chamfering (edge grinding) produces a
process-induced degradation layer on the edge. Top (surface)
grinding induces another process-induced degradation layer on the
top surface. The process-induced degradation layer is a surface
layer having disorder of the lattice structure. Plenty of
dislocations are included in the process-induced degradation layer.
The process-induced degradation layer does not mean the inclusion
of impurities. It is undesirable for mirror wafers to remain the
process-induced degradation layer on the top or bottom surfaces.
The top and bottom process-induced degradation layers should be
removed by etching. Instead of the top/bottom surfaces, the present
invention observes circumference edge grinding (chamfering,
bevelling) closely. Chamfering of the prevalent semiconductor
wafers, e.g., silicon wafers and GaAs wafers employs resin-bonding
or metal-bonding whetstones which bond diamond granules with a
resin or a metal to the base. The prevailing resin-bonding
whetstones and the metal-bonding whetstones are acute, hard and
sturdy. Chamfering time is short. However, diversion of the
prevalent resin- or metal-bonding whetstones to GaN chamfering
invites breaks and cracks on the GaN wafers at high rates,
generates thick process-induced degradation layers and induces
large bow. GaN is easily damaged by shock because of high rigidity,
low toughness and high fragility. The high crack occurrence rate
originates from the fact that the resin- or metal-bonding
whetstones have granules rigidly fixed to the bases and the bonding
materials cannot absorb the shocks acting on fragile wafer edges by
the granules. Frequent cracks and breaks occur on GaN wafers by
chamfering with metal- or resin-bonding whetstones. Even in the
case of free from break of crack, 20 .mu.m-50 .mu.m thick
process-induced degradation layers M and large bow are induced by
the metal- or resin-bonding whetstone chamfering. Sometimes the bow
curvature radius is less than 1 m. The prevalent resin- or
metal-bonding whetstones are inapplicable to chamfering of the
highly rigid and highly fragile GaN wafers.
[0030] A first purpose of the present invention is to provide a
method of chamfering a nitride wafer without breaking and cracking.
A second purpose of the present invention is to provide a method of
chamfering a nitride wafer without inducing bow. A third purpose of
the present invention is to provide a method of chamfering a
nitride wafer with a controlled thickness of an edge
process-induced degradation layer. A fourth purpose is to provide a
method of chamfering a nitride wafer with a high wafer processing
yield. A fifth purpose is to provide a method of chamfering a
nitride wafer with a high device production yields. A sixth purpose
is to provide a nitride wafer with a high device production yield.
Outer portions of top and bottom surfaces are called "peripheries"
in the present description. The circular side which is
perpendicular to the top/bottom is called a "circumference". A
crossing circle between the top/bottom and the circumference is
called a "ridge". A set of a periphery, a ridge and a circumference
is called an "edge".
DISCLOSURE OF THE INVENTION
[0031] A nitride semiconductor wafer of the present invention is
made by grinding and etching a bottom surface of an as-grown
nitride substrate, chamfering the wafer by a whetstone bonded with
diamond granules or diamond/oxide granules by a soft bonding
material, making an edge process-induced degradation layer of a 0.5
.mu.m-10 .mu.m thickness, more favorably, a 1 .mu.m-3 .mu.m
thickness, grinding, polishing and etching a top surface. The soft
bonding material means rubbers or bubbled resins. Sizes of diamond
granules are #3000 to #600. Mixing of oxide granules into diamond
granules enables the whetstone to chamfer wafers by mechanical
action of diamond and chemical action of oxides. Cooperation of the
diamond mechanical action and the oxide chemical action is called
"mechanochemical effect". Pertinent oxide granules to the
chamfering whetstones are Fe.sub.2O.sub.3, CuO, MnO.sub.2,
Cr.sub.2O.sub.3, ZnO and so on. The whetstones are endowed with
elasticity. Sometimes it will take much time for elastic whetstones
to chamfer nitride wafers in the present invention.
[0032] The present invention chamfers nitride wafers by a
soft-material bonded whetstone with a base bonded with fine diamond
granules and oxide granules. Chamfering by the soft whetstone
alleviates shocks acting on edges of wafers, prevents the wafer
from breaking and cracking, decreases the thickness of an edge
process-induced degradation layer, suppresses bow. The present
invention employs rubber-bonding or bubbled resin-bonding
whetstones for chamfering in order to reduce mechanical shock to
the edge. Chamfering causes an edge process-induced degradation
layer. The thickness d of the edge process-induced degradation
layer M is a measure of the shocks acting on edges of wafers during
chamfering. Grosser diamond granules and harder bonding materials
induce larger shocks on edges. In this case, postchamfered
observation of edges shows a thick edge process-induced degradation
layer. Finer diamond granules and softer bonding materials cause
smaller shocks of edges. In the latter case, postchamfered
observation of edges reveals a thin edge process-induced
degradation layer.
[0033] Chamfering wafers by hard whetstones with a base bonded with
gross diamond (low mesh numbers #) by metals or hard resins invites
a large edge process-induced degradation layer with a thickness d
of 20 .mu.m-50 .mu.m. In this case, excess strong shocks induce
breaks and cracks of wafers. The present invention succeeds in
suppressing occurrences of breaks and cracks by gently chamfering
wafers with elastic, soft whetstones and keeping a thickness d of
the edge process-induced degradation layer in the range of 0.5
.mu.m-10 .mu.m. The present invention employs rubbers or bubbled
resins as bonding materials, chooses fine diamond granules (high
mesh numbers #) and adds oxide granules to the diamond granules for
making soft whetstones. The use of the soft whetstones enables the
present invention to chamfer GaN wafers without causing cracks,
breaks and bow. Chamfering GaN wafers with a whetstone having a
pertinent elasticity gives a 0.5 .mu.m-10 .mu.m thick edge
process-induced degradation layer. The edge process-induced
degradation layer in the range of d=0.5 .mu.m-10 .mu.m thicknesses
suppresses bow and increases a device yield.
[0034] FIG. 1 demonstrates the whole of planar-processing steps
starting from an as-grown freestanding GaN wafer. An as-grown GaN
wafer is processed as shown in FIG. 1. Bottom surface (N-plane)
grinding (A) is done for adjusting the thickness to a predetermined
value. The bottom grinding (A) induces a thick process-induced
degradation layer M on the bottom surface. The process-induced
degradation layer is a layer having a deformed lattice structure.
There are plenty of dislocations in the process-induced degradation
layer. The CL (cathode luminescense) can discern the
process-induced degradation layer from other regions. The CL
observation can measure the thickness d of the process-induced
degradation layer. The process-induced degradation layer is not a
layer including impurities. Since the process-induced degradation
layer includes deformation of lattice structure, it is undesirable
to remain the process-induced degradation layer on the bottom
surface. The bottom process-induced degradation layer should be
removed. The GaN wafer is wet-etched (B) by hot KOH, NaOH or
H.sub.3PO.sub.4 solution. Wet etching (B) by hot KOH, NaOH or
H.sub.3PO.sub.4 solution eliminates the process-induced degradation
layer from the bottom. The bottom grinding can be replaced by
bottom-polishing.
[0035] The edge of the wafer is ground (chamfer; C) by an elastic
whetstone having a base bonded by a rubber or a bubbled resin with
fine diamond granules or diamond/oxide mixed granules. The edge
grinding is called "chamfering". Chamfering prevents the wafer from
cracking or breaking.
[0036] The present invention recommends rubber whetstones for
chamfering GaN wafers. The rubber whetstone is a whetstone having a
base bonded by rubber with stationary whetgranules. The
whetgranules are hard granules, e.g., diamond granules. Since the
bonding material is rubber, the whetstone should be precisely
called "rubber-bonding whetstone" or "rubber-bond whetstone". But
in this description the whetstone is sometimes called "rubber
whetstone" for short. This name does not mean that granules are
rubber. The present invention first advocates the use of rubber
whetstones for chamfering GaN wafers.
[0037] After chamfering (C), the top surface of the wafer is ground
(D) and polished (E). Top grinding (D) causes a 10 .mu.m-50 .mu.m
thick process-induced degradation layer M on the top surface. Top
polishing (E) induces another 0.1 .mu.m-20 .mu.m thick
process-induced degradation layer M on the top. Wafers for
epitaxial growth require smooth tops without process-induced
degradation layers. For the purpose of making a smooth surface on
the top, the top of the postgrinding wafer is polished. Grinding
(D) and polishing (E) produce the top process-induced degradation
layers. The top is dry-etched in vapor phase (F) till the top
process-induced degradation layer M is fully eliminated. The wafer
processing includes the bottom grinding (A), bottom etching (B),
chamfering (C), top grinding (D)/polishing (E) and top
vapor-phase-etching (F). The contrivance of the present invention
resides neither in the bottom grinding (A) nor top grinding
(D)/polishing (E). The purpose of the present invention is an
improvement of chamfering (C).
[0038] The gist of the present invention is nitride wafer
chamfering of grinding an edge of a nitride wafer by a soft, rubber
or bubbled resin whetstone bonded with fine diamond granules or
diamond/oxide granules and keeping an edge process-induced
degradation layer of 0.5 .mu.m-10 .mu.m thicknesses, more
favorably, of 1 .mu.m-3 .mu.m thicknesses. The resultant 0.5
.mu.m-10 .mu.m thick edge process-induced degradation layers at the
edge can prevent the wafer from cracking and can suppress bow. In
the following steps, the resultant 0.5 .mu.m-10 .mu.m edge
process-induced degradation layers can prohibit the wafer from
chipping and can raise device production yield.
[0039] The following is further details of a series of steps of the
wafer processings of the present invention. Grinding invites a
process-induced degradation layer M with a perturbed lattice
structure on a ground surface of a wafer. Larger stationary
granules of the whetstone and a heavier load on the surface invite
a thicker process-induced degradation layer M. On the contrary,
finer stationary granules and a lighter load on the surface cause a
thinner process-induced degradation layer M.
[0040] As shown in FIG. 1, the wafer planar-processing includes a
series of the steps of bottom grinding (A), bottom etching (B),
chamfer (C), top grinding (D), top polishing (E) and top etching
(F). The bottom grinding (A) causes a thick bottom process-induced
degradation layer M having a thickness of 10 .mu.m-50 .mu.m, which
depends upon conditions of grinding. It is undesirable to leave a
thick process-induced degradation layer M on the bottom. The bottom
process-induced degradation layer is removed by wet-etching the
wafer in a hot KOH, NaOH or H.sub.3PO.sub.4 solution.
[0041] An N-plane (nitrogen) differs from a Ga-plane (gallium) in a
C-plane GaN wafer. A bottom surface (Nitrogen-plane; N-plane;
(000-1)) is chemically weaker than a top surface (Gallium-plane;
Ga-plane; (0001)). One of KOH, NaOH and H.sub.3PO.sub.4 solutions
can etch the bottom surface of the GaN wafer. It is available to
use some other acids or alkalis capable of etching the bottom
surface. The top surface (Ga-plane) is chemically far stronger than
the bottom surface (N-plane). None of KOH, NaOH and H.sub.3PO.sub.4
solutions can etch the top surface of the GaN wafer. Top and bottom
surfaces require different etching methods for removing
process-induced degradation layers.
[0042] Then an edge of the GaN wafer is ground (chamfer C). The GaN
wafer is chamfered by rotating an elastic, rubber or bubbled-resin
bonding whetstone bonded with whetgranules, bringing the rotating
whetstone in contact with the edge of the wafer, rotating the wafer
and grinding the whole of the edge. The stationary granules of the
whetstone are fine diamond granules or diamond/oxide mixture
granules. Mesh numbers (#) of the diamond granules bonded to the
whetstone by rubber or bubbled resin are #600 (20 .mu.m diameter)
to #3000 (4 .mu.m diameter). The mesh number # is a measure
signifying an average diameter of whetgranules. A larger mesh
number # signifies smaller granules. A smaller mesh number #
indicates grosser granules. Chamfering produces a blunt edge and
causes a process-induced degradation layer on the edge. The present
invention controls a thickness d of the edge process-induced
degradation layer within the range of 0.5 .mu.m-10 .mu.m, more
favorably within the range of 1 .mu.m-3 .mu.m. Chamfering by a hard
whetstone with larger granules (smaller #) induces a thick edge
process-induced degradation layer of 20 .mu.m-50 .mu.m and large
bow and sometimes invites cracks or breaks. Chamfering by a
whetstone with fine diamond granules can decrease the thickness d
of the edge process-induced degradation layer. However decrement of
sizes of granules is still insufficient. The present invention
proposes an addition of oxide granules to the whetstone, which is
effective in alleviating the strong mechanical action of diamond
granules. Pertinent oxides are Fe.sub.2O.sub.3, Cr.sub.2O.sub.3,
MnO.sub.2, CuO, ZnO, Fe.sub.3O.sub.4 and so on which are chemically
unstable oxides. The present invention has discovered first that
some kinds of metal oxide granules have a chemical faculty of
facilitating grinding of nitride wafers. The present invention
names the chemical faculty "mechanochemical" effect. Grinding by a
whetstone with diamond/oxide granules is named "mechanochemical
grinding". Edge grinding by a diamond/oxide whetstone is called
"mechanochemical chamfering".
[0043] The mechanochemical grinding softens the surfaces of nitride
wafers and decreases the grinding load by oxidization reaction by
oxide granules. The mechanochemical chamfering can suppress
occurrence of process-induced degradation layers and cracks during
chamfer.
[0044] All oxides are not applicable to the present invention.
Stable oxides are inappropriate. SiO.sub.2 and Al.sub.2O.sub.3 are
unsuitable, because SiO.sub.2 and Al.sub.2O.sub.3 are stable oxides
and lack the mechanochemical effect. Addition of stable oxide
granules cannot decrease the thickness d of the edge
process-induced degradation layer below 10 .mu.m. The bonding
material which fixes whetgranules on a base is soft rubber or
bubbled resin endowed with elasticity. A complementary assembly of
a high mesh number of diamond granules, an addition of oxide
granules and an elastic bonding material enables chamfering to
lower the edge process-induced degradation layer thickness d to 10
.mu.m or less (d.ltoreq.10 .mu.m). The present invention proposes
the use of rubber whetstones or bubbled resin bonding whetstones
for the nitride wafer chamfer for the first time. The present
invention proposes the use of the whetstones of diamond/oxide
mixing granules for the nitride wafer chamfer for the first
time.
[0045] After chamfering, the top surface is ground. A
process-induced degradation layer occurs on the top surface. Since
devices will be fabricated on the top surface, the top is a
mirror-flat surface. Then the top surface is polished. Polishing of
the top contains gross polishing by gross free granules and fine
polishing by fine free granules. For example, 30 .mu.m-3 .mu.m
diameter free granules can be used for the gross polishing. 3
.mu.m-0.1 .mu.m diameter free granules can be applied to the fine
polishing. Polishing produces a new process-induced degradation
layer on the top. Top grinding and top polishing are
planar-processings which are usually done in series for raising
productivity and enhancing quality. Sometimes one of grinding and
polishing can be omitted for reducing cost. Top grinding is done by
whetstones with stationary granules of e.g., #3000 to #8000. Top
polishing is done by free granules of 30 .mu.m-15 .mu.m diameters.
It is possible to replace the top polishing by grinding using a
whetstone with a high mesh number # or to replace the top grinding
by policing using rough granules. In this case, the top grinding
uses a whetstone with a mesh number of #3000 to #8000 and the top
polishing uses whetstone granules within the range of 30 .mu.m-15
.mu.m.
[0046] At the step, there are process-induced degradation layers on
the top and the bevelled edge. Vapor phase etching on the top
surface eliminates only the top surface process-induced degradation
layer. Chemical-mechanical polishing (CMP) can remove the top
surface process-induced degradation layer in stead of the vapor
phase etching. A process-induced degradation layer of a thickness
between 0.5 .mu.m and 10 .mu.m, preferably between 1 .mu.m and 3
.mu.m remains on the bevelled edge. Unlike flat top surface or flat
bottom surface, it is difficult for etching to adjust the thickness
of the process-induced degradation layer on the bevelled edge.
There is another series of wafer processing. The series includes
the steps of slicing of a nitride crystal ingot into as-cut wafers,
bottom surface etching (B), chamfering (C), top surface polishing
(E) and vapor phase etching (F). The alternative processing can
dispense with the bottom surface grinding (A) and top surface
grinding (D), since the slicing can adjust the thicknesses of the
as-cut wafers. The alternative processing requires the bottom
surface etching (B) and top surface etching (F), since the slicing
induces process-induced degradation layers on both top and bottom
surfaces. For example, the alternative processing has a series of
slicing.fwdarw.bottom surface etching (B).fwdarw.vapor phase
etching (F).fwdarw.chamfering (C).fwdarw.top surface polishing
(E).fwdarw.vapor phase etching (F). A single vapor phase etching
can replace the double vapor phase etching steps. CMP can
substitute for the second vapor phase etching. In an alternate case
of making a tall GaN ingot and slicing the ingot into as-cut
wafers, the side of the ingot can be ground first and the ingot is
sliced into wafers next. This case can dispense with the first
bottom grinding (A) for adjusting wafer thicknesses.
[0047] This invention chamfers nitride wafers by a soft
rubber-bonding or bubbled resin bonding whetstone embedded with
fine diamond whetgranules or oxide granules in addition to diamond
whetgranules. The use of the soft whetstone can suppress breaks or
cracks of wafers. The 0.5 .mu.m-10 .mu.m thick, preferably 1
.mu.m-3 .mu.m thick process-induced degradation layers remaining on
the bevelled edge enables us to obtain small bow nitride wafers or
bow-free nitride wafers. Crack occurrence rate is low and wafer
processing yield is high on the nitride wafers of the present
invention. Further this invention can enhance the device production
yield of fabricated on the nitride wafers.
[0048] The thickness d of the edge process-induced degradation
layer is a measure of judging the strength of mechanical shock
acting on wafers at chamfering afterward. The edge process-induced
degradation layer having a thickness from 0.5 .mu.m to 10 .mu.m
signifies a soft contact and a weak shock to the wafer edge by the
whetstone.
BRIEF DESCRIPTION OF THE DRAWINGS
[0049] FIG. 1 is a process diagram showing the steps of making
mirror wafers from an as-grown circular nitride crystal grown on an
undersubstrate in vapor phase and separated from the
undersubstrate.
[0050] FIG. 2 is a graph showing a relation between the thickness d
of peripheral process-induced degradation layers and the bow U of
the GaN wafers. The abscissa is the logarithmically represented
thickness d (.mu.m) of the peripheral process-induced degradation
layers. The ordinate is the bow U (.mu.m).
[0051] FIG. 3 is a graph showing a relation between the measured
thickness d of peripheral process-induced degradation layers and
the wafer production yield Y (solid line) and the crack occurrence
rate C (broken line). The abscissa is the logarithmically
represented thickness d (.mu.m) of peripheral process-induced
degradation layers. The righthand ordinate means the crack
occurrence rate C(%). The lefthand ordinate is the substrate
production yield Y(%).
[0052] FIG. 4 is a graph showing a relation between the thickness d
of peripheral process-induced degradation layers and the on-wafer
device yield Q. The abscissa is logarithmically-represented
thickness d (.mu.m) of peripheral process-induced degradation
layers. The ordinate indicates the on-wafer device yield Q(%).
[0053] FIG. 5 is a sectional view of a present invention GaN wafer
having a peripheral process-induced degradation layer of a
thickness between 0.5 .mu.m and 10 .mu.m, preferably between 1
.mu.m and 3 .mu.m on an upper round edge (including a
circumference).
[0054] FIG. 6 is a sectional view of another present invention GaN
wafer having a peripheral process-induced degradation layer of a
thickness between 0.5 .mu.m and 10 .mu.m, preferably between 1
.mu.m and 3 .mu.m on a round edge (including a circumference).
[0055] FIG. 7 is a sectional view of another present invention GaN
wafer having a peripheral process-induced degradation layer of a
thickness between 0.5 .mu.m and 10 .mu.m, preferably between 1
.mu.m and 3 .mu.m on an upper bevelled edge.
[0056] FIG. 8 is a sectional view of another present invention GaN
wafer having a peripheral process-induced degradation layer of a
thickness between 0.5 .mu.m and 10 .mu.m, preferably between 1
.mu.m and 3 .mu.m on edges and a circumference.
[0057] FIG. 9 is a sectional view showing a chamfering process of
shaping a sharp edge (including a circumference) of a wafer into a
round edge by a rubber-bonding whetstone.
[0058] FIG. 10 is a sectional view showing a chamfering process
which shapes an edge (including a circumference) of a wafer into a
slanting edge by a rubber-bonding whetstone.
[0059] FIG. 11 is a perspective view of a chamfering apparatus,
which was proposed by Japanese patent laying open No. 2004-319951
(Japanese Patent application No. 2003-275935), for chamfering an
edge of a wafer by carrying a whettape in contact with a wide range
of the edge of the wafer and rotating the wafer.
[0060] FIG. 12 is a sectional view of a contacting part of the
chamfering apparatus, which was proposed by Japanese patent laying
open No. 2004-319951 (Japanese Patent application No. 2003-275935),
for chamfering an edge of a wafer by carrying a whettape in contact
with a wide range of the edge of the wafer and rotating the
wafer.
[0061] FIG. 13 is a sectional view of a bow wafer with the
periphery laid on a horizontal plane for explaining the definition
of the degree of bow by the height U of the center of the wafer W
of a diameter D.
[0062] FIG. 14 is a graph showing a relation between the thickness
d (.mu.m) of the process-induced degradation layers and the wafer
bow U (.mu.m) of Samples 1 to 11. Numerals attached denote sample
numbers. .smallcircle. denotes an allowable sample and x indicates
a rejected sample (same hereinafter).
[0063] FIG. 15 is a graph showing a relation between the thickness
d (.mu.m) of the process-induced degradation layers and the crack
occurrence rate C(%) of Samples 1 to 11. Numerals attached denote
sample numbers.
[0064] FIG. 16 is a graph showing a relation between the thickness
d (.mu.m) of the process-induced degradation layers and the
device-production yield Q(%) of Samples 1 to 11. Numerals attached
denote sample numbers.
[0065] FIG. 17 is a graph showing a relation between the oxide
whetgranule ratio (wt %) and the wafer edge oxygen concentration
(at %) of Samples 12 to 18. The oxide whetgranule ratio+diamond
whetgranule ratio=100%.
[0066] FIG. 18 is a graph showing a relation between the oxide
whetgranule ratio (wt %) and the edge process-induced degradation
layer thickness d (.mu.m) of Samples 12 to 18.
[0067] FIG. 19 is a graph showing a relation between the oxide
whetgranule ratio (wt %) and the chipping occurrence ratio p (%) of
Samples 12 to 18.
[0068] FIG. 20 is a graph showing a relation between the oxide
whetgranule ratio (wt %) and the device production yield Q(%) of
Samples 12 to 18.
[0069] FIG. 21 is a graph showing a relation between the edge
process-induced degradation layer thickness d (.mu.m) and the edge
metal concentration m (at %) of Samples 19 to 26.
[0070] FIG. 22 is a graph showing a relation between the edge
process-induced degradation layer thickness d (.mu.m) and the
chipping occurrence rate p(%) of Samples 19 to 26.
[0071] FIG. 23 is a graph showing a relation between the edge
process-induced degradation layer thickness d (.mu.m) and the
device production yield Q (%) of Samples 19 to 26.
[0072] FIG. 24 is a graph showing a relation between the edge
process-induced degradation layer thickness d (.mu.m) and edge
roughness Ra (.mu.m) of Samples 1 to 26.
[0073] FIG. 25 is a sectional view of an epitaxial wafer having a
gallium nitride substrate prepared by the present invention, an
n-GaN film, an n-AlGaN film, a light emission layer, a p-AlGaN film
and a p-GaN film epitaxially grown the gallium nitride substrate
for fabricating light emitting devices.
[0074] FIG. 26 is a sectional view of a unit of a device having a
gallium nitride substrate prepared by the present invention, an
n-GaN film, an n-AlGaN film, a light emission layer, a p-AlGaN film
and a p-GaN film epitaxially grown on the gallium nitride
substrate, a p-electrode and an n-electrode.
[0075] FIG. 27 is a sectional view of a light emitting device which
is made by fixing the device chip in an epi-down posture (p-region
down and n-region up) on a stem of a package with AuSn solder and
connecting the n-electrode with a leadpin by a wire.
[0076] FIG. 28 is a graph showing a relation between the edge
oxygen concentration O (at %) and the chipping occurrence rate p
(%) of Samples 12 to 18.
[0077] FIG. 29 is a graph showing a relation between the edge
oxygen concentration O (at %) and the device production yield Q(%)
of Samples 12 to 18.
[0078] FIG. 30 is a graph showing a relation between the edge metal
concentration m (at %) and the chipping occurrence rate p (%) of
Samples 19 to 26.
[0079] FIG. 31 is a graph showing a relation between the edge metal
concentration m (at %) and the device production yield Q (%) of
Samples 19 to 26.
BEST FEATURES FOR BRINGING THE INVENTION INTO PRACTICE
[0080] Nitride semiconductor wafers are composed of group III
nitride, e.g., GaN, AlN, InN, AlGaN, InGaN and so on. An HVPE
method, a flux method and an ammonothermal method can produce GaN
crystals. An HVPE method, a sublimation method and a flux method
can grow AlN crystals. An InN crystal can be grown by an HVPE
method. Slicing a grown nitride crystal by a wiresaw or a bladesaw
produces as-cut wafers. Planar-processing includes grinding and
polishing. Etching includes dry-etching and wet-etching.
[0081] The present invention can chamfer wide, thick GaN wafers,
e.g., of a 5 inch (12.7 cm) diameter and a 850 .mu.m thickness by
rubber-bonding whetstones and porous resin-bonding whetstones.
Non-porous resin bonding whetstones, metal bonding whetstones and
electrodeposition whetstones, which are too hard, are impertinent
to chamfer nitride semiconductor wafers. A pertinent material of
the bonding rubber is chloroprene rubber (CR). CR is excellent in
elasticity, in-process elastic deformation and removability. A
suitable range of rubber hardness is low hardness of 40 to 60 for
suppressing cracks and damage from occurring in chamfering.
[0082] The whetstone is a porous resin-bonding whetstone with a
porosity of 20% to 50%. Porous bonding-resin is made by mixing
calcium carbonate (CaCO.sub.3) or other carbonate with a material
resin having whetgranules and sintering the mixture. Calcium
carbonate emits gas bubbles in the resin during the sintering. Gas
bubbles gives the resin porosity. Polyvinyl alcohol (PVA) or phenol
resin is a suitable resin for bonding whetgranules on a whetstone
disc.
[0083] Whetgranules to be bonded on the whetstone disc are diamond
(C) granules. Alumina (Al.sub.2O.sub.3) granules, silicon carbide
(SiC) granules or boron nitride (BN) granules can replace the
diamond whetgranules. Diamond, alumina, silicon carbide and boron
nitride granules have a similar mechanical whetting function. Thus
diamond whetstones will be explained as a typical example
chamfering whetstone hereafter. Sizes of whetgranules are
represented by mesh (#). The present invention uses #220 to #6000
diamond granules as stationary whetgranules bonded on bases of
chamfering whetstones. The mesh (#) is a measure indicating sizes
of granules. Higher number of mesh (#) signifies smaller granules.
Lower number of mesh (#) means bigger granules.
[0084] A larger-granule bonded whetstone can chamfer an edge of a
wafer in a shorter time. But the use of the larger-granule bonded
whetstone induces more cracks and breaks and larger bow. Use of a
smaller-granule bonded whetstone wastes much time. Sometimes the
smaller-granule whetstone invites more bow and a higher crack
occurrence rate. The rate of the stationary granules in the bonding
material should be 3 vol % to 20 vol %. The symbol "vol %" means
volume %.
[0085] An allowable range of the edge process-induced degradation
layer thickness d of a nitride wafer is 0.5 .mu.m to 10 .mu.m. Less
than 0.5 .mu.m or more than 10 .mu.m of the edge process-induced
degradation layer thickness d causes large bow and high rates of
crack occurrence. 1 .mu.m-3 .mu.m of thickness is more favorable
for the edge process-induced degradation layer. Wafer bow is
sensitive to the edge process-induced degradation layer, inner
stress and the top/bottom process-induced degradation layers. The
bow depends also upon the diameter and thickness of a wafer. CL
(cathode luminescense) and TEM (transmission electron microscope)
of cleavage planes can observe and estimate the process-induced
degradation layer.
[0086] A preferable range of roughness of the chamfered edge is
Ra0.07 .mu.m to Ra3 .mu.m in arithmetic average roughness Ra. Ra is
one of representations of roughness. Ra is obtained by calculating
an average height, summing absolute values of the differences
between the heights of measuring spots and the average height and
averaging the difference absolute values. Rms, which is another
roughness representation, is obtained by summing squares of the
differences between the heights of measuring points and the average
height, averaging the sum and taking a root of the average. A same
surface gives different values to Ra and Rms for roughness.
Irregularity of surfaces varies relations between Ra and Rms. Ra is
neither equal nor proportional to Rms. Don't confuse Ra with Rms.
Edge roughness smaller than Ra0.07 .mu.m or larger than Ra3 .mu.m,
which is unfavorable, increases the impurity concentration on the
chamfered edges. The impurities on the edges originate from the
bonding material of the whetstone and the wax supporting the wafer
to the holder. A more favorable range of edge roughness is Ra0.15
.mu.m-Ra1 .mu.m. Edge roughness is observed and measured in 30
.mu.m square (30 .mu.m.times.30 .mu.m) fields by a laser microscope
with laser diodes of a wavelength of 658 nm
[0087] Whetstones of excess high mesh numbers of diamond granules
can safely chamfer gallium nitride wafers. Diamond whetstones
consisting only of low mesh numbers of diamond granules fail to
chamfer. Too strong mechanical action of the low mesh diamond
whetstones induces ruggedness of edges of gallium nitride wafers.
Use of a complex granule whetstone including a mixture of diamond
whetgranules and oxide whetgranules is effective in lowering the
roughness of chamfered edges. Unstable oxides should be chosen as
oxide whetgranules for inviting chemical reactions. The chemical
reactions help chamfering. The addition of the chemical reactions
of oxide granules enables the whetstone to reduce the necessary
amount of diamond granules, the mechanical action due to diamond
granules and physical shocks. Favorable oxide granules are
Fe.sub.2O.sub.3, Fe.sub.3O.sub.4, Cr.sub.2O.sub.3, CuO,
CO.sub.3O.sub.4, MnO.sub.2 and ZnO granules. The oxide granules
have a chemical function and a mechanical function. The functions
are here named "mechanochemical" effect. Mechanochemical effect is
a new effect. The mechanochemical effect of the oxide whetgranules
assists diamond granules in eliminating superficial parts and
succeeds in chamfering with a load (shock) far lighter than the
case of chamfering by diamond granules alone. The mechanochemical
effect enables chamfering to produce smooth and scar-free edges.
Stable oxides, e.g., silica (SiO.sub.2) and alumina
(Al.sub.2O.sub.3), which have no mechanochemical effect, are
inappropriate. (0001) planes (Ga-plane; top surface) of nitride
wafers are highly stable and chemically resistant. The
mechanochemical effect would fail to grind the (0001) plane. The
object of chamfering is less stable, chemically-weak planes
inclining to the stable (0001) plane. Thus the whetstones including
oxide granules succeed in mechanochemical grinding of edges
(chamfering) of nitride wafers with high efficiency.
[0088] Grinding apparatus ambience and oxide granules discharge
oxygen atoms. A part of the discharged oxygen atoms invades the
chamfered edges of wafers. 3 at % to 20 at % are favorable oxygen
concentrations in the chamfered edges. Less than 3 at % of edge
oxygen concentration induces chipping of wafers. More than 20 at %
of edge oxygen concentration causes increment of the rate of
rejected devices fabricated on a peripheral part of the processed
wafer. The edge oxygen concentration can be controlled by the
mixing ratio of oxide whetgranules bonded to the base of the
whetstone. Further the edge oxygen concentration can be adjusted by
regulating the oxidization effect of the washing liquid after
chamfering. More favorable range of the edge oxygen concentration
is 5 at %-15 at %. Atomic components in the edges of nitride wafers
can be measured by AES (Auger Electron Spectroscopy).
[0089] A variety of semiconductor devices can be fabricated on the
hitherto mentioned nitride wafers as a substrate. For example,
light emitting diodes are made by a series of the following steps.
An epitaxial wafer is produced by growing a 1 .mu.m thick n-GaN
layer, a 150 nm thick n-Al.sub.0.1Ga.sub.0.9N layer, a light
emission layer, a 20 nm thick p-Al.sub.0.2Ga.sub.0.8N layer and a
150 nm thick p-GaN layer in order epitaxially on the n-type GaN
wafer made by the present invention. The light emission layer is
composed of a multiquantum well structure having four 10 nm thick
GaN shield films and three 3 nm thick Ga.sub.0.85In.sub.0.15N well
films alternately piled (FIG. 25).
[0090] An n-electrode is formed on the bottom (N-plane; (000-1)
plane) n-GaN substrate. A p-electrode is formed on the top p-GaN
layer. Plenty of devices are fabricated on the n-GaN wafer. For
example, the wafer is cut into LED (light emitting diode) chips of
2 mm squares. Device chips are mounted upon stems of packages. In
the case of epi-up mounting, wherein the bottom of the substrate is
fitted on the stem, the n-electrode should be wide and the
p-electrode should be narrow. Light emitted from the emission layer
goes out via the p-region. In the other case of epi-down mounting,
wherein the top of the p-layer is fitted on the stem, the
n-electrode should be narrow and the p-electrode should be wide
(FIG. 26). Light emitted from the emission layer goes out via the
n-region. FIG. 27 demonstrates an epi-down mounted LED which can
emit light through the n-GaN substrate, since the GaN substrate is
transparent. The epi-down mounting device excels in thermal
diffusion, because the emission layer is close to the stem metal
which carries heat away downward quickly. The GaN substrate is
endowed with high conductivity. Heat diffusion is helped by the
high thermal conductivity of the GaN substrate. The thermal
conductivity of GaN is far higher than sapphire. On-GaN devices are
suitable for producing high power devices.
Embodiment 1
[1. Production of GaN Wafer of a Variety of Thicknesses of Edge
Process-Induced Degradation Layer]
[0091] A plurality of circular 2 inch (about 50 mm) diameter GaN
as-cut wafers are produced by a vapor phase growth method. The GaN
wafers are n-type wafers doped with silicon (Si) or oxygen (O). The
GaN wafers are processed to mirror wafers by a series of
planar-processings shown in FIG. 1.
A: Bottom surface grinding B: KOH bottom surface etching C:
Chamfering by rubber-bonding whetstones of mesh numbers of #600 to
#3000 D: Top surface grinding E: Top surface polishing (gross and
minute) F: Vapor phase etching of top surface A: The bottom surface
grinding is a planar process of reducing thicknesses of wafers W to
a predetermined range of thickness. Grinding of a bottom surface
(N-plane) decreases the thickness of the wafer. The grinding causes
a process-induced degradation layer M with lattice disorder on the
bottom surface. Neither human eyesight nor scanning electron
microscope (SEM) can discern the process-induced degradation layer
M. Cathode luminescence (CL), fluorescence microscope or
transmission electron microscope (TEM) can discriminate the
process-induced degradation layer M from the background. The
process-induced degradation layer M is a superficial region which
is not illuminant when a GaN wafer is examined by cathode
luminescence (CL) or a fluorescence microscope. Alternatively the
process-induced degradation layer M is a superficial region which
is darker than neighboring deeper region in a light vision of a
transparent electron microscope (TEM). The thickness d of the
process-induced degradation layer M can be measured by the CL,
fluorescence microscope or TEM. For example, a definite region of a
GaN wafer is observed by the SEM and CL at the same time. There is
a CL-non-illuminant region near the surface which has been planar
processed. The depth of the CL-non-illuminant region is the
thickness of the process-induced degradation layer. The SEM can
exactly measure the depth of the CL-non-illuminant region. The
planar processing causes a thick process-induced degradation layer
M of a thickness of d=10 .mu.m-50 .mu.m to the bottom surface. B:
It is undesirable to remain the process-induced degradation layer
M. The bottom surface process-induced degradation layer M should be
eliminated. Step B is to remove the process-induced degradation
layer M induced on the bottom surface. Step B etches the wafer in a
25% KOH solution at 90.degree. C. for 10 minutes to 120 minutes.
Etching depth is controlled within the range of 1 .mu.m to 50 .mu.m
by changing the etching time. C: Chamfering is a step of bevelling
the circumference edge into a slop or a round. Chamfering by a
resin-bonding whetstone induces a break or a crack originating from
the periphery. Use of the whettape as proposed by Japanese Patent
Laying Open No. 2004-319951 will waste too much time. This
invention employs a rubber-bonding whetstone for chamfering in
order to protect the circumference from breaking or cracking.
[0092] A rubber-bonding whetstone fixes stationary whetgranules
with a rubber onto a base disc. The whetstone chamfers the sharp
edge of a wafer into an oblique or round edge by coming into
contact with the edge and rotating around a center.
[0093] FIG. 9 demonstrates a chamfering process in which a round
pulley-shaped rubber-bonding whetstone is chamfering the edge E
(including a circumference) of a wafer W into a round section.
[0094] FIG. 10 shows another chamfering process in which a
cone-shaped rubber-bonding whetstone H having a slanting
cone-surface is bevelling the edge E of a wafer into a slanting
edge. In this case, both upper edge and lower edge shall be
bevelled. Slanting bevelled edges are formed. A middle part of the
circumference is not ground in the case. A variety of shapes other
than pulley-shape and cone-shape are available for the whetstone.
For example, a moving small whetstone can chamfer a wafer by moving
around the wafer.
[0095] Preferably wafers are chamfered by 600 mesh rubber-bonding
whetstone at a first step. Then 2-000 mesh rubber-bonding whetstone
shall be used at a second step. Chamfering by the 600 mesh
whetstone gives roughness about Ra0.3 .mu.m to the edge E.
Chamfering by the fine 200 mesh whetstone improve the edge
roughness to about Ra0.07 .mu.m. The edge becomes smooth.
Chamfering causes a process-induced degradation layer M on the edge
E.
D: Top surface grinding D reduces a wafer with a thickness of about
600 .mu.m-550 .mu.m to 500 .mu.m-520 .mu.m. The grinding amount
(thickness loss) can be controlled by grinding time. For instance,
it takes about 60 minutes to grind a top surface of a 580 .mu.m
thick wafer into a 510 .mu.m thick. A process-induced degradation
layer M is induced on the top surface of the wafer W by the top
surface grinding. E: Top surface polishing comprises gross
polishing and minute polishing of a top surface. Two steps of
polishing (gross and minute) gives the top surface flatness. There
are two process-induced degradation layers M on both the top
surface and the edge. The thicknesses d of the top surface and the
edge process-induced degradation layers M vary according to the
grinding or polishing amounts. 10 .mu.m to 50 .mu.m are in general
the range of thicknesses of the process-induced degradation layers
induced on the top and the edge. F: The top surface process-induced
degradation layer can be removed by mechanical-chemical polishing
(MCP), electrolysis polishing, liquid phase etching or vapor phase
etching. In the present case, the top process-induced degradation
layer is eliminated by vapor phase etching with a heated gas
including chlorine (Cl.sub.2). The top surface (Ga-plane) is
chemically and physically more refractory than the bottom surface
(N-plane). Alkali etching is useless for corroding the top. The
vapor phase etching is employed for removing the top surface
process-induced degradation layer.
[0096] A variety of wafers having different thicknesses d of the
edge process-induced degradation layers M are prepared by changing
the conditions of grinding and etching. The edge process-induced
degradation layer thicknesses of the GaN wafers are approximately
d=0 .mu.m to 20 .mu.m.
[2. Relation Between the Edge Process-Induced Degradation Layer
Thickness and the Bow of the Wafer]
[0097] There are some different expressions for the bow of wafers.
One expression is a curvature radius R of a wafer central line.
Another is a curvature rate 1/R. Another is a central height of a
wafer. In the present invention, the bow is expressed by the
central height U of a wafer which is placed with the periphery
contacting to a flat plane.
[0098] FIG. 13 demonstrates the definition of the bow U which is a
height of the center from the flat plane. A relation U=D.sup.2/8 R
connects the curvature radius R with the central height U, where D
is a diameter of the wafer and R is a curvature radius. The bows U
of the sample wafers and edge process-induced degradation layer
thicknesses d are measured.
[0099] The relation between the edge process-induced degradation
layer thickness d and the wafer bow U is studied. FIG. 2 exhibits
the result. The abscissa denotes logarithms of the edge
process-induced degradation layer thickness d (.mu.m). The ordinate
denotes the bow U (.mu.m).
[0100] Although there are other factors affecting bow, FIG. 2
indicates that the bow U of wafers can be expressed by a monovalue
function of the edge process-induced degradation layer thickness d.
The bow U monotonously increases between d=3 .mu.m and d=20 .mu.m,
as shown in FIG. 2. d=10 .mu.m gives about U=50 .mu.m. Thicknesses
d=0.5 .mu.m-10 .mu.m of edge process-induced degradation layers
give wafers bow U less than or equal to 50 .mu.m. A point within
d=3 .mu.m-1 .mu.m gives wafers the minimum bow U. Decrease of the
edge process-induced degradation layer thicknesses d under d=1
.mu.m raises the bow U.
[3. Relation Between the Edge Process-Induced Degradation Layer
Thickness d and the Crack Occurrence Rate C]
[0101] Crack occurrence rates C are measured for the sample wafers.
FIG. 3 denotes the result of measuring the crack occurrence rates
C. The abscissa of FIG. 3 denotes logarithm of the edge
process-induced degradation layer thickness d (.mu.m). The right
ordinate denotes the crack occurrence rate C (%). A broken curve
shows the crack occurrence rate. Various factors cause cracks. This
result teaches that the crack occurrence rate C has a distinct
relation to the edge process-induced degradation layer thickness
d.
[0102] The crack occurrence rate C(%) increases monotonously with
the edge process-induced degradation layer thickness d in the range
of d=3 .mu.m to d=10 .mu.m. A decrease from d=1 .mu.m to d=0 .mu.m
raises the crack occurrence rate C. d=0.5 .mu.m-10 .mu.m can
suppress the crack occurrence rate less than or equal to 50%. A
value of d below 3 .mu.m minimizes the crack occurrence rate C.
[4. Relation Between the Edge Process-Induced Degradation Layer d
and the Wafer Production Yield Y]
[0103] The wafer production yield Y(%) is examined. Although a
variety of factors affect the wafer production yield, the wafer
production yield Y has a clear relation to the edge process-induced
degradation layer thickness d.
[0104] In FIG. 3, a solid curve denotes the wafer production yield
(Y) as a function of the edge process-induced degradation layer
thickness d. The wafer production yield Y is approximately 80% to
75% in the range of d=3 .mu.m-10 .mu.m. Y is about 72% to 80% in
the range of d=0.5 .mu.m-10 .mu.m. Y is 76% at d=10 .mu.m. Y for
d=10 .mu.m-20 .mu.m is not depicted, since Y falls further.
[0105] At d=1 .mu.m of the edge process-induced degradation layer
thickness, the wafer production yield Y decreases to about 75%. At
d=0 .mu.m, Y falls to about 60%. 0.5 .mu.m-10 .mu.m edge
process-induced degradation layer thicknesses maintain the wafer
production yield Y at 72% to 80%.
[5. Relation Between the Edge Process-Induced Degradation Layer
Thickness d and the in-Surface Device Production Yield Q]
[0106] Light emitting devices are produced on the GaN wafers
processed by a variety of conditions. In-surface device production
yields Q are examined. It turned out that the in-surface device
production yield Q also has a strong relation to the edge
process-induced degradation layer thickness d.
[0107] FIG. 4 exhibits the result of measurements. The edge
process-induced degradation layer thickness d=1.7 .mu.m gives the
maximum of 80% to the in-surface device production yield Q. The
thickness d exceeding 3 .mu.m decreases the device production yield
Q. On the contrary, the process-induced degradation layer thickness
d approaching 0 .mu.m lowers the device production yield Q via 50%
below 40%. More than or equal to 60% of device production yield Q
requires a 0.5 .mu.m-10 .mu.m thickness of the edge process-induced
degradation layer. 1 .mu.m-5 .mu.m thicknesses d of the edge
process-induced degradation layer can maintain the in-surface
device production yield of more than or equal to 70%.
[0108] Since an edge process-induced degradation layer contains
lattice disorder, one may assume that d=0 .mu.m would be the best.
The above results deny the assumption. Edge process-induced
degradation layer thicknesses of d=0.5 .mu.m-10 .mu.m are
favorable, in particular, d=1 .mu.m-3 .mu.m are preferable.
[0109] The optimum range of edge process-induced degradation layer
thickness d is 1 .mu.m to 3 .mu.m in consideration of all the
factors of the bow U, the wafer production yield Y, the crack
occurrence rate C and the in-surface device production yield Q.
[0110] The above experiment mentions a series of bottom surface
grinding, bottom surface etching, edge chamfering, top surface
grinding, top surface polishing and top surface vapor phase etching
of 2 inch (50 mm) diameter GaN wafers which have been obtained as a
single as-cut wafer in a batch of growth. Similar results are
obtained in another case of a plurality of GaN wafers made in a
batch of growth by growing a tall GaN crystal and slicing the GaN
into a plurality of as-cut wafers. The alternative case includes
another series of bottom etching, edge chamfering, top surface
polishing and top surface vapor phase etching.
[0111] The above are results of experiments for GaN wafers. Other
nitride semiconductor wafers of AlN, AlGaN or InGaN wafers reveal
similar results. Hereafter "surface" succeeding "top" or "bottom"
is sometimes omitted for simplicity. "Bottom surface" or "top
surface" will often be briefly written "bottom" and "top".
[0112] On the accumulation of results of experiments mentioned
above, the present invention proposes a method of giving a 0.5
.mu.m-10 .mu.m thickness, preferably, 1 .mu.m-3 .mu.m thickness of
a process-induced degradation layer to a circumference edge of a
nitride wafer by chamfering the edge with a rubber-bonding or a
bubble-resin-bonding whetstone in a series of processes of bottom
grinding, edge chamfering, and top grind/polishing.
[0113] FIGS. 5 to 8 show sectional views of nitride semiconductor
wafers W processed by the above mentioned methods.
[0114] FIG. 5 exhibits a nitride wafer W having a 0.5 .mu.m-10
.mu.m, preferably 1 .mu.m-3 .mu.m thick process-induced degradation
layer M at an upper edge which was prepared by chamfering an edge
by a pulley-shaped rubber-bonding whetstone (G), polishing the top
and eliminating a top process-induced degradation layer.
[0115] FIG. 6 exhibits a nitride wafer W having a 0.5 .mu.m-10
.mu.m, preferably 1 .mu.m-3 .mu.m thick process-induced degradation
layer M at a round edge (including a side) which was prepared by
chamfering an edge by a pulley-shaped rubber-bonding whetstone (G),
polishing the top and eliminating a top process-induced degradation
layer.
[0116] FIG. 7 exhibits a nitride wafer W having a 0.5 .mu.m-10
.mu.m, preferably 1 .mu.m-3 .mu.m thick process-induced degradation
layer M at a slant edge which was prepared by chamfering an edge by
a cone-shaped rubber-bonding whetstone (G), polishing the top and
eliminating a top process-induced degradation layer.
[0117] FIG. 8 exhibits a nitride wafer W having a 0.5 .mu.m-10
.mu.m, preferably 1 .mu.m-3 .mu.m thick process-induced degradation
layer M at slant edges (including a side) which was prepared by
chamfering edges and a side by a cone-shaped rubber-bonding
whetstone (G), polishing the top and eliminating a top
process-induced degradation layer.
Embodiment 2
Embodiment 2
Samples 1-11, Table 1; FIGS. 14, 15, 16
[0118] Embodiment 1 uses 2 inch (50 mm) diameter wafers. Increase
of a diameter of a wafer raises the difficulty of chamfering. A
larger diameter of wafers induces more cracks, greater bow and
lower device production yield. The usefulness of the present
invention will rise, if this invention is applicable to large
diameter wafers.
[0119] Embodiments 2 to 4 describe 26 samples which give one step
of chamfering to large and thick GaN wafers with a 5 inch (127 mm)
diameter and an 850 .mu.m thickness for confirming the
applicability of the present invention to wide wafers. There is no
antecedent of a 5 inch diameter GaN wafer. The present invention
proposes 5 inch wide GaN wafers for the first time. Sample numbers
are indicated by affixed numerals in FIGS. 14 to 24. .smallcircle.
means an allowable, passable sample. x means a rejected sample. The
processes of gallium nitride wafers which are common to all Samples
1-26 are explained.
[Processes of Gallium Nitride Wafers]
[0120] FIG. 1 demonstrates a series of processes of bottom surface
grinding (A), KOH etching (B) for bottom process-induced
degradation layer removal (B), chamfering (C), top surface grinding
(D), top surface polishing (E) and vapor phase etching for top
surface process-induced degradation layer removal. Chamfering (C)
characterizes the present invention. All the processes (A) to (F)
are explained.
(1) Shaping and Processing of GaN Wafers
[0121] An oxygen doped GaN ingot of a 5 inch (127 mm) diameter is
grown by the HVPE method. A wire-saw cuts the GaN ingot in a (0001)
plane into a 850 .mu.m thick GaN substrate wafer. The bottom
surface of the GaN wafer is ground by a #600 diamond whetstone
(process A). The bottom grinding induces a process-induced
degeneration layer on the bottom surface of the wafer. The
process-induced degeneration layer is eliminated from the bottom by
a wet etching in a 15% KOH solution at 50.degree. C. (process B).
Instead of KOH, NaOH, H.sub.3PO.sub.4 or another acid or alkali
solution is available for the bottom etching (B) as long as the
etching speed is high enough. Dry etching is also useful to
eliminate the process-induced degeneration layer from the bottom.
After the etching, a variety of whetstones listed on Table 1
chamfer the GaN wafers (process C). Then top surfaces of the wafers
are processed (processes D, E).
(2) Lapping and CMP (Chemical Mechanical Polishing) of a Top
Surface of a GaN Wafer
[0122] A bottom surface (N-plane; (000-1) plane) of an n-type GaN
wafer (or another nitride wafer) is glued with wax to a ceramic
holder. A turntable of a 600 mm diameter is mounted on a lapping
apparatus. The holder is laid upon the turntable. Slurry dispersed
with free diamond granules is supplied via a slurry inlet onto the
turntable. The turntable is revolved around a central axis. A
weight is laid on the holder. The holder is rotated around a shaft.
The top surface (Ga-plane; (0001) plane) is lapped by pushing the
top surface to the turntable, rotating the holder and revolving the
turntable.
[0123] A copper (Cu) turntable or a tin (Sn) turntable is employed
for lapping top surfaces. Three different sizes of free diamond
granules having a diameter of 9 .mu.m, 3 .mu.m and 2 .mu.m
respectively are prepared. The diameters of granules are decreased
step by step according as the lapping proceeds. Polishing pressure
is 100 g/cm.sup.2 to 500 g/cm.sup.2. Both the rotation of the
holder (GaN wafer) and the revolution of the turntable are 30
turns/min to 60 turns/min. The lapping produces mirror flatness on
the top surface.
[0124] Polishing follows lapping. A polishing pad is pasted on a
turntable of a 600 mm diameter of a polishing apparatus. The holder
with the lapped GaN wafer is laid on the polishing pad. A weight
pushes the holder to the polishing pad. Slurry dispersed with free
whetgranules is supplied via a slurry inlet to the polishing pad of
the polishing apparatus. The turntable revolves around a center
shaft. The holder rotates around a center. CMP (chemical mechanical
polishing) of the top surface (Ga-plane; (0001) plane) of the GaN
wafer is done by pushing the top surface to the pad, rotating the
holder and revolving the turntable.
[0125] Slurry is prepared by dispersing 2 .mu.m diameter alumina
(Al.sub.2O.sub.3) granules at 5 wt % in water, adding HNO.sub.3 and
adjusting a pH within pH2-pH4. The polishing pads are polyurethane
suede pads. The turntable is a stainless-steel turntable. Polishing
pressure is 50 g/cm.sup.2 to 600 g/cm.sup.2. Rotation speeds of
both the GaN wafer and the polishing pad are 30 turns/min to 70
turns/min.
(3) Dry-Etching and Washing of Top Surfaces of GaN Wafers
[0126] Following the CMP processing, the GaN wafers are dry-etched
in an ICP-RIE apparatus. The etching gas is chlorine gas. The
dry-etched GaN wafers are washed with a dilute KOH solution and IPA
(isopropyl alcohol). The finished GaN wafers are tested. Testing
items are edge process-induced degradation layer, top/edge/bottom
roughnesses, bow, edge oxygen concentration, edge metal
concentration and chipping occurrence. Top surface roughnesses are
commonly Ra1.5 nm for all samples. Bottom surface roughnesses are
commonly Ra5.6 .mu.m for all samples. Edge roughnesses are varied
for samples. Tables 1, 2 and 3 show the results of measurements of
the parameters.
(4) Fabrication of LED Devices on the n-GaN Substrate
[0127] The prepared n-GaN wafers have 1.times.10.sup.-2 .OMEGA.cm
of resistivity and 3.times.10.sup.18/cm.sup.3 of carrier
concentration. The n-GaN wafers are installed in an MOCVD
apparatus. The MOCVD apparatus produces epitaxial GaN wafers by
growing a 1 .mu.m thick Si-doped n-GaN layer, a 150 nm thick
Si-doped n-Al.sub.0.1Ga.sub.0.9N layer, a light emission layer, a
20 nm thick Mg-doped p-Al.sub.0.2Ga.sub.0.8N layer and a 150 nm
thick Mg-doped p-GaN layer in turn on the top surface (Ga-surface:
(0001) plane) of the n-GaN wafer. FIG. 25 shows a sectional view of
the epitaxial wafer. The light emission layer is a multiquantum
well consisting of four 10 nm thick GaN shield layers and three 3
nm thick Ga.sub.0.85In.sub.0.15N well layers alternately piled.
[0128] Then the bottom surface ((000-1) plane, N-plane) is provided
with a 100 .mu.m diameter n-electrode by piling a 200 nm thick Ti
layer, a 1000 nm thick Al layer, a 200 nm thick Ti layer and a 2000
nm thick Au layer on the bottom surface and annealing the wafer in
nitrogen gas atmosphere. The top p-GaN layer is provided with a
p-electrode by piling a 4 nm thick Ni layer and a 4 nm thick Au
layer and annealing the wafer in inert gas atmosphere. FIG. 26
shows a section of unit device of the wafer. The wafer is divided
into plenty of 2 mm square device chips. The p-electrode of a chip
is epi-down bonded with AuSn solder on a stem of a package. The
n-electrode is connected with a leadpin by a wire. FIG. 27 shows a
semiconductor device having an LED structure made by the above
processes.
[0129] The epi-down mounted LED has advantages over epi-up mounted
LED. The epi-down structure enables the LED to emit light via the
transparent GaN substrate. The epi-down structure excels in thermal
diffusion, since heat can flow down from the light emission layer
via the close stem to the package. GaN has excellent heat
conductivity of 210 W/mK as five times as much as sapphire of 40
W/mK. The epi-down structure allows the GaN substrate to guide and
diffuse heat upward. The epi-down LED can make the best use of the
features of GaN wafers. Light power is measured by laying an LED at
a center of an integral sphere and supplying the LED with a 4 A
injection current. The injection of current makes the emission
layer emit light. A part of light upward goes via the substrate out
of the LED. Another part of the light is emitted downward, is
reflected by the stem and is thrown upward out of the LED (FIG.
27). Another part of light which is reflected by sides of the
package goes outward. The LED has a very large size of a 2 mm
square. The epi-down mounting allows even downward emitted light to
go out upward. The present invention can give large output power
LEDs made on the GaN substrate.
[0130] The output of light emitting devices is measured by placing
the LEDs at a central point in an integral sphere, injecting a
current of 4 A to the LED for inducing the LED for emitting light,
gathering light by the integral sphere, sensing the power gathered
by a detector placed on another center. High light power is one of
the most significant performances of LEDs. The LEDs emitting light
power of 2 W or more are judged "allowable". The LEDs of light
output of less than 2 W are judged "rejected". The critical output
is 2 W. The above mentioned planar-processings, device fabrication,
inspection are common to all Samples 1-26.
[0131] Samples 1-11 examine bow, crack occurrence and device
production yield by changing diamond mesh numbers, bonding
materials, kinds of oxides and rates of oxides of chamfering
whetstones.
TABLE-US-00001 TABLE 1 Sample Sample Sample Sample Sample Sample
Sample Sample Sample Sample Sample 1 2 3 4 5 6 7 8 9 10 11 Edge
Bond- Kinds Rubber Rubber Rubber Rubber Rubber Resin Rubber Resin
Metal Metal Electro- Pro- ing deposi- cessing tion Materials CR CR
CR CR CR PVA CR Phenol Cu--Sn Cu--Sn Ni Rubber Hardness 40 40 40 45
50 -- 60 -- -- -- -- Porosity (vol %) 0% 0% 0% 0% 0% 40% 0% 0% 0%
0% 0% Whet- Diamond(Mesh No.) 6000 3000 3000 2000 1500 800 600 600
600 3000 3000 granule Oxide Fe.sub.2O.sub.3 Fe.sub.2O.sub.3 -- --
-- -- -- -- -- Fe.sub.2O.sub.3 -- Oxide Amount(wt %) 20 20 -- -- --
-- -- -- -- 20 -- Wafer Edge process-induced degradation 0.3 0.5 1
2 3 6 10 13 18 13 16 Prop- layer thickness d (.mu.m) erties Edge
roughness Ra (.mu.m) 0.03 0.07 0.15 0.5 1 3 5 7 8 1 2 Bow U (.mu.m)
20 10 7 6 5 5 12 25 50 28 43 Crack occurrence rate C (%) 9 4 2 0 1
2 5 12 26 14 22 Device Device production yield Q (%) 52 72 82 84 85
80 70 50 20 52 35
Table 1 Samples 1-11; mesh numbers, bonding materials, oxides, bow,
edge roughness, crack occurrence, device production yields
[0132] The first line of Table 1 denotes sample numbers. Sample
numbers are 1-11. Wafer properties are values measured for each
sample. Conditions of chamfering, wafer properties and device
properties are shown for all samples.
[0133] The second to eighth denote the properties of chamfering
whetstones. Chamfering is an edge grinding process by rotating a
whetstone embedded with stationary whetgranules and bringing an
edge of a wafer into contact with the rotating whetstone. "Bond"
indicates the properties of the bonding materials which fix
whetgranules to a base. Prevalent whetstones are resin-bonding
whetstones which maintain stationary granules by a resin to the
base, metal-bonding whetstones which fix granules by a metal to the
base and electrodeposition whetstones which fix granules to the
base by electrodeposition. The present invention denies these
whetstones. The present invention does not use resin-bonding
whetstones, metal-bonding whetstone and electrodeposition
whetstones. Rigid fixation of granules to the base is a drawback of
the prevailing resin-, metal-bonding or electrodeposition
whetstones.
[0134] This invention chamfers edges of nitride wafers by rubber
bonding whetstones or bubbled-resin bonding whetstones. For the
purpose of comparison, nitride wafers are chamfered by resin
bonding whetstones, metal bonding whetstones and electrodeposition
whetstones. The column of materials denotes the bonding materials.
"CR" means chloroprene rubber. The next column shows the hardness
of the bonding rubbers. Bonding resins or bonding metals are harder
than the bonding rubbers. Measurements of bonding metals or resins
are different from the measurement of the bonding rubber. Resins
are still poor in elasticity and inapplicable for nitride wafer
chamfering. Bubbling brings resins to be useful for nitride wafer
chamfering by enhancing elasticity. Bubbling of resin makes a
mixture of resin and air bubbles. The "porosity" is the volume
percent of air bubbles to the whole. The porosity is a measure of
the elasticity of the bubbled resin.
[0135] Table 1 shows the kinds and weight percent of oxide whet
granules. 100% minus oxide whet granule weight percent is the
weight percent of diamond whet granules.
[0136] The lines of substrate characteristics denote four
characteristics of the postchamfered wafers. The first one is the
thicknesses (.mu.m) of the process-induced degeneration layers at
edges of the wafers. The cathode luminescence (CL) can discern the
process-induced degeneration layers from the other parts. The CL
measures the thicknesses of the process-induced degeneration layers
remaining at the periphery of the wafers. The second line denotes
the roughness Ra of the chamfered peripheries in .mu.m unit. The
third line shows the bow of the wafers. The bow is represented by
the central height U of an wafer which is placed with the periphery
in contact with a flat plane. Since the present invention treats
with wide wafers, even a small bow curvature shows a big bow height
U. The bow height U of a wafer of a diameter D is related to the
bow curvature radius R by quadratic function approximation
R=D.sup.2/8 U. For example, the central height U=20 .mu.m for a
wafer of a diameter D=127 mm corresponds to a curvature radius
R=100 m of bow. The curvature of bow is 1/R=0.01 m.sup.-1.
[0137] "Crack" means occurrence of breaks or splits during
chamfering. "Crack occurrence rate" is a ratio of the number of
wafers which are broken or split while chamfered to the number of
total wafers. Samples 1-11 are chamfered at the edges under a
variety of conditions. Bow, crack occurrence rates, device yields
are measured for the postchamfered samples 1-11.
[0138] The "device" line shows device production yields which are
ratios of the number of good device chips to the number of the
whole device chips obtained from one of the sample wafers. The
device is a very large-sized light emitting diode (LED) of a 2 mm
square (4 mm.sup.2 area). Conventional LEDs on sale have a 300
.mu.m square (0.09 mm.sup.2 area) to a 500 .mu.m square (0.25
mm.sup.2 area). The LEDs of the present invention are fifteen-forty
times as large as the conventional ones. A 127 mm diameter circular
wafer has about a 12000 mm.sup.2 area which is about 3000 times as
wide as the 2 mm square chips. Thus about 3000 LEDs are obtained
from a single 127 mm diameter wafer. A predetermined number of LEDs
are chosen from a batch of LEDs produced on one GaN wafer as
samples. Qualities of the LEDs are tested by injecting currents.
Since the LEDs of the above samples have very wide areas (e.g., 2
mm square), the probability of occurrence of rejected devices would
be assumed to become higher than conventional tiny devices (0.3 mm
square to 0.5 mm square) from a statistical standpoint. This
invention, however, turns out to be able to produce good
large-sized LEDs with high yields.
[0139] FIG. 14 is a graph denoting a relation between measured edge
process-induced degradation layer thicknesses d (.mu.m) and wafer
bow U (.mu.m) of Samples 1 to 11. .smallcircle. denotes an allowed
sample. x denotes a rejected sample. Numerals affixed are sample
numbers. The present invention advocates a range denoted by an
arrow between 0.5 .mu.m and 10 .mu.m as desirable edge
process-induced degradation layer thicknesses. A more favorable
range is d=1 .mu.m-3 .mu.m. FIG. 15 denotes a relation between
measured edge process-induced degradation layer thicknesses d
(.mu.m) and crack occurrence rates C(%).
[0140] FIG. 16 denotes a relation between measured edge
process-induced degradation layer thicknesses d (.mu.m) and device
production yields Q (%).
[Sample 1 (Fe.sub.2O.sub.3; Rubber Whetstones; Q=52%)]
[0141] 5 inch (127 mm) diameter GaN circular wafers of a 850 .mu.m
thickness are planar-processed by bottom surface grinding (A) and
bottom surface etching (B). The GaN wafers are chamfered by a
rubber bonding whetstone (C) with chloroprene rubber (CR) as a
bonding material. Since the bonding material is rubber, the
whetstone should be called "rubber-bonding whetstone". But the
whetstone is sometimes called a "rubber whetstone" in brief by
omitting "bonding". Furthermore what grinds objects is not rubber
but diamond granules. Rubber, a bonding material, cannot be
bubbled. Thus porosity of the rubber whetstones is 0%. Resin,
another bonding material, can be bubbled. Rubber has different
degrees of hardness by varying components. This rubber of the
whetstone has hardness 40.
[0142] Stationary whetgranules are a mixture of diamond granules
and ferric oxide (Fe.sub.2O.sub.3) granules. The ratio of the oxide
granules (Fe.sub.2O.sub.3) is 20 wt %. The rest of 80% is diamond
granules. The mesh number of diamond granules is #6000. The mesh
number denotes the average size of granules. A larger mesh number
means a smaller diameter of granules.
[0143] #6000 denotes fine diamond granules of an average diameter
of 2.5 .mu.m. The thickness d of the edge process-induced
degradation layer originating from the chamfering is d=0.3 .mu.m.
Edge roughness of the postchamfered wafers is Ra0.03 .mu.m. The
postchamfered GaN wafers are further planar-processed by top
surface grinding (D) and top surface polishing (E). The top surface
becomes mirror flat. The top process-induced degradation layer
caused by the processes (D) and (E) is eliminated by top surface
etching (F). The bow height U is 20 .mu.m. The crack occurrence
rate C is 9%. Crack means occurrence of crack, split or break of a
wafer while chamfering. Crack occurrence rate C (%) is a ratio
obtained by dividing the number of the wafers cracking, splitting
or breaking while chamfering by the total number of the wafers and
multiplying with 100. The last line "device yield" Q is obtained by
producing 2 mm square LED devices on the GaN wafer, dividing the
wafer into LED chips, mounting LED chips on stems, wirebonding
electrodes to leadpins, injecting a 4 A current to LEDs, examining
the light output, counting the number of approved LEDs of over 2 W
output and dividing the number of the approved LEDs by the total
number of tested LEDs. The device production yield Q of Sample 1 is
52%. Poor production yield degrades Sample 1. Sample 1 should be
rejected. In Figures Sample 1 is affixed with "x" which denotes
rejected one. It is assumed that too small edge process-induced
degradation layer thickness d=0.3 .mu.m would reduce the production
yield Q. Thicker edge process-induced degradation layers are
desired. Use of grosser whetstones or application of heavier loads
will increase the edge process-induced degradation layer thickness
d.
[Sample 2 (Fe.sub.2O.sub.3; Rubber Whetstones; d=0.5 .mu.m;
Q=72%)]
[0144] After bottom grinding (A) and bottom etching (B), 5 inch
(127 mm) diameter GaN circular wafers of a 850 .mu.m thickness are
chamfered by a rubber bonding whetstone (C) with chloroprene rubber
(CR) as a bonding material. The porosity of the rubber-bonding
whetstones is 0%. This rubber of the whetstone has hardness 40.
Stationary whetgranules are a mixture of diamond granules and
ferric oxide (Fe.sub.2O.sub.3) granules. The ratio of the oxide
granules (Fe.sub.2O.sub.3) is 20 wt %. The rest of 80% is diamond
granules. The mesh number of diamond granules is #3000 which is
grosser than Sample 1. The average size of #3000 whetgranules is
about 5 .mu.m. The thickness d of the edge process-induced
degradation layer originating from the chamfering is d=0.5 .mu.m.
Use of a grosser whetstone increases the edge process-induced
degradation layer thickness d. Edge roughness of the postchamfered
wafers is Ra0.07 .mu.m. The grosser whetstone raises the edge
roughness. The postchamfered GaN wafers are further
planar-processed by top surface grinding (D) and top surface
polishing (E). The top surface becomes mirror flat. The top
process-induced degradation layer is eliminated by top surface
etching (F). The bow height U, which is a height from a center to
the periphery, is 10 .mu.m. Sample 2 is superior to Sample 1 in bow
height U. The bow curvature radius is R=200 m. The crack occurrence
rate C is 4% which is less than half of Sample 1. The device
production yield Q which is obtained by producing 2 mm square LED
devices on the wafer, dividing the wafer into LED chips, mounting
LED chips on stems, wirebonding electrodes to leadpins, injecting a
4 A current to LEDs, examining the light output, counting the
number of approved LEDs of over 2 W output and dividing the number
of the approved LEDs by the total number of tested LEDs, is 72%.
High production yield approves Sample 2 useful. Sample 2 is
allowable. In Figures Sample 2 is affixed with ".smallcircle."
which denotes an approved one. In the following figures,
.smallcircle. is approved and x is rejected. Edge roughness of
Sample 2 is Ra0.07 .mu.m which is rougher than Sample 1. The edge
process-induced degradation layer thickness d is 0.5 .mu.m which is
thicker than Sample 1. The high yield of Sample 2 results from a
pertinent edge roughness and a suitable edge process-induced
degradation layer thickness d.
[Sample 3 (No Oxide Granules; Rubber Whetstones; d=1 .mu.m;
Q=82%)]
[0145] After bottom grinding (A) and bottom etching (B), 5 inch
(127 mm) diameter GaN circular wafers of a 850 .mu.m thickness are
chamfered by a rubber bonding whetstone (C) with chloroprene rubber
(CR) as a bonding material. The porosity of the rubber-bonding
whetstones is 0%. This rubber of the whetstone has hardness 40.
Stationary whetgranules are all diamond granules. No oxide granules
are used. The ratio of the oxide granules is 0 wt %. The whole is
diamond granules. The mesh number of diamond granules is #3000
which is equal to Sample 2. The thickness d of the edge
process-induced degradation layer originating from the chamfering
is d=1 .mu.m Increase of the mechanical grinding function by 100%
diamond granules increases the edge process-induced degradation
layer thickness d. Edge roughness of the postchamfered wafers is
Ra0.15 .mu.m. The 100% diamond granule whetstone raises the edge
roughness. The postchamfered GaN wafers are further
planar-processed by top surface grinding (D) and top surface
polishing (E). The top surface becomes mirror flat. The top
process-induced degradation layer is eliminated by top surface
etching (F). The bow height U is 7 .mu.m, which is smaller than
Samples 1 and 2. The bow curvature radius is R=286 m. 1 .mu.m
thickness of the edge process-induced degradation layer reduces the
bow. The crack occurrence rate C is 2% which is less than Samples 1
and 2. The device production yield Q which is obtained by producing
2 mm square LED devices on the wafer, is 82%. Sample 2 is
allowable. The 1 .mu.m thickness of the edge process-induced
degradation layer, which is bigger than Samples 1 and 2, improves
all of the bow U, crack occurrence C, and device production yields
Q in Sample 3.
[Sample 4 (No Oxide Granules; Rubber Whetstones; d=2 .mu.m;
Q=84%)]
[0146] After bottom grinding (A) and bottom etching (B), 5 inch
(127 mm) diameter GaN circular wafers of a 850 .mu.m thickness are
chamfered by a rubber bonding whetstone (C) with chloroprene rubber
(CR) as a bonding material. The porosity of the rubber-bonding
whetstones is 0%. This rubber of the whetstone has hardness 45,
which is a little harder than Samples 1,2 and 3. Stationary
whetgranules are all diamond granules. No oxide granules are used.
The ratio of the oxide granules is 0 wt %. The whole is diamond
granules. The mesh number of diamond granules is #2000 (average
diameter 7-8 .mu.m), which is grosser than Samples 1, 2 and 3. The
thickness d of the edge process-induced degradation layer
originating from the chamfering is d=2 .mu.m. Increase of the
mechanical grinding function by grosser diamond granules increases
the edge process-induced degradation layer thickness d. Edge
roughness of the postchamfered wafers is Ra0.5 .mu.m. The grosser
diamond granule whetstone raises the edge roughness. The
postchamfered GaN wafers are further planar-processed by top
surface grinding (D) and top surface polishing (E). The top surface
becomes mirror flat. The top process-induced degradation layer is
eliminated by top surface etching (F). The bow height U is 6 .mu.m.
The bow curvature radius is R=333 m, which is smaller than Samples
1, 2 and 3. 2 .mu.m thickness of the edge process-induced
degradation layer reduces the bow. The crack occurrence rate C is
0% which is less than Samples 1, 2 and 3. The edge process-induced
degradation layer thickness d=2 .mu.m gives the minimum crack
occurrence rate C in all samples. The device production yield Q by
producing 2 mm square LED devices on the wafer like the preceding
samples, is 84%. Sample 4 is allowable. The d=2 .mu.m thickness of
the edge process-induced degradation layer, which is bigger than
Samples 1, 2 and 3, improves all of the bow U, crack occurrence C,
and device production yields Q in Sample 4. The d=2 .mu.m of the
edge process-induced degradation layer thickness gives the optimum
performance of all characteristics.
[Sample 5 (No Oxide Granules; Rubber Whetstones; d=3 .mu.m;
Q=85%)]
[0147] After bottom grinding (A) and bottom etching (B), 5 inch
(127 mm) diameter GaN circular wafers of a 850 .mu.m thickness are
chamfered by a rubber bonding whetstone (C) with chloroprene rubber
(CR) as a bonding material. The porosity of the rubber-bonding
whetstones is 0%. This rubber of the whetstone has hardness 50,
which is a little harder than Samples 1-4. Stationary whetgranules
are all diamond granules. No oxide granules are used. The mesh
number of diamond granules is #1500 (average diameter 10 .mu.m),
which is grosser than Samples 1-4. The thickness d of the edge
process-induced degradation layer originating from the chamfering
is d=3 .mu.m. Increase of the mechanical grinding function by
grosser diamond granules increases the edge process-induced
degradation layer thickness d. Edge roughness of the postchamfered
wafers is Ra1 .mu.m. The grosser diamond granule whetstone raises
the edge roughness. The postchamfered GaN wafers are further
planar-processed by top surface grinding (D) and top surface
polishing (E). The top surface becomes mirror flat. The top
process-induced degradation layer is eliminated by top surface
etching (F). The bow height U is 5 .mu.m, which is the smallest in
Samples 1-5. The bow curvature radius is R=400 m. 3 .mu.m thickness
of the edge process-induced degradation layer reduces the bow. The
crack occurrence rate C is 1%, which is less than Samples 1, 2 and
3. The d=3 .mu.m of the edge process-induced degradation layer
thickness gives a good result. The device production yield Q by
producing 2 mm square LED devices on the wafer like the preceding
samples, is Q=85%, which is the best in all Samples 1-11. Sample 5
is allowable. The d=3 .mu.m thickness of the edge process-induced
degradation layer is a favorable value for bow suppression, crack
suppression and device yield.
[Sample 6 (No Oxide Granules; Resin-Bonding Whetstones; d=6 .mu.m;
Q=80%)]
[0148] After bottom grinding (A) and bottom etching (B), 5 inch
(127 mm) diameter GaN circular wafers of a 850 .mu.m thickness are
chamfered by a resin-bonding whetstone (C) with polyvinyl alcohol
(PVA) as a bonding material. This is not a rubber-bonding whetstone
but a resin-bonding whetstone, since the bonding material is a
resin (PVA). Pure resins are useless due to hardness. The hardness
of the resin bonding material is reduced by bubbling PVA resin.
Bubbling lowers hardness of a resin. The porosity of the
resin-bonding whetstones is 40%. High porosity gives the resin
bonding material pertinent elasticity. The hardness of resin cannot
be measured by the same method as rubber. Stationary whetgranules
are all diamond granules. No oxide granules are used. The mesh
number of diamond granules is #800 (average diameter 19 .mu.m),
which is grosser than Samples 1-5. The thickness d of the edge
process-induced degradation layer originating from the chamfering
is d=6 .mu.m, which is twice larger than Sample 5. Increase of the
mechanical function by grosser diamond granules increases the edge
process-induced degradation layer thickness d. Edge roughness of
the postchamfered wafers is Ra3 .mu.m. The grosser diamond granule
whetstone raises the edge roughness. The postchamfered GaN wafers
are further planar-processed by top surface grinding (D) and top
surface polishing (E). The top surface becomes mirror flat. The top
process-induced degradation layer is eliminated by top surface
etching (F). The bow height U is 5 .mu.m (R=400 m), which is equal
to Sample 5 and is the smallest in Samples 1-6. The minimum of bow
U is given by d=4-5 .mu.m. Since d of Sample 6 is thicker than
Samples 1-4, bow is smaller than Samples 1-4. The crack occurrence
rate C is 2%, which is smaller than Samples 1-2, larger than
Samples 4-5 and equal to Sample 3. As well as rubber-whetstones,
bubbled resin bonding whetstones which are endowed with elasticity
by bubbling can suppress the occurrence of cracks. The device
production yield Q by producing 2 mm square LED devices on the
wafer like the preceding samples, is Q=80%. It is a high device
yield. Sample 6 is allowable. The results of Samples 1-6 suggest
that d=1 .mu.m-3 .mu.m gives an optimum condition for crack
suppression, bow suppression and device yield enhancement. In
particular, d=2 .mu.m of the edge process-induced degradation layer
thickness is the best value for giving strong crack suppression and
maximum device yield.
[Sample 7 (No Oxide Granules; Rubber Whetstones; d=10 .mu.m;
Q=70%)]
[0149] After bottom grinding (A) and bottom etching (B), 5 inch
(127 mm) diameter GaN circular wafers of a 850 .mu.m thickness are
chamfered by a rubber-bonding whetstone (C) with chloroprene rubber
(CR) as a bonding material. The porosity of the rubber-bonding
whetstone is 0%. The CR rubber has hardness 60, which is harder
than Samples 1-5. Stationary whetgranules are all diamond granules.
No oxide granules are used. The mesh number of diamond granules is
#600 (average diameter 25 .mu.m), which is grosser than Samples
1-6. The thickness d of the edge process-induced degradation layer
originating from the chamfering is d=10 .mu.m which is the thickest
among Samples 1-7. The mechanical function by grosser diamond
granules increases the edge process-induced degradation layer
thickness d. Edge roughness of the postchamfered wafers is Ra5
.mu.m. The grosser diamond granule whetstone raises the edge
roughness. The postchamfered GaN wafers are further
planar-processed by top surface grinding (D) and top surface
polishing (E). The top surface becomes mirror flat. The top
process-induced degradation layer is eliminated by top surface
etching (F). The bow height U is 12 .mu.m (R=170 m), which is
larger than Samples 2-6. Increment of the edge process-induced
degradation layer thickness d raises the bow U from Sample 6. The
crack occurrence rate C is 5%, which is smaller than Sample 1 but
larger than Samples 2-6. In a range of d<2 .mu.m, a decrease of
d raises the crack occurrence rates C and an increase of d lowers
C. The crack occurrence rate C takes the minimum at d=2 .mu.m. In a
range of d>2 .mu.m, increment of d raises the crack occurrence
rate C. The device production yield Q by producing 2 mm square LED
devices on the wafer like the preceding samples, is Q=70%. The
large d=10 .mu.m lowers the device production yield Q in Sample 7.
When 82% or more of the device production yield Q is required
(Q.gtoreq.0.82 .mu.m), the edge process-induced degradation layer
thickness d should be 1 .mu.m.ltoreq.d.ltoreq.3 .mu.m. When 70% or
more of Q is required, an allowable range of d is expanded to 0.5
.mu.m.ltoreq.d.ltoreq.10 .mu.m. In this case of 0.5
.mu.m.ltoreq.d.ltoreq.10 .mu.m, Sample 1 is rejected (x) and
Samples 2-7 are allowable (.smallcircle.). Samples 2-5, and 7 teach
that hardness 40-60 is an optimum range of rubber hardness of the
rubber whetstones.
[Sample 8 (No Oxide Granules; Resin-Bond Whetstones; d=13 .mu.m;
Q=50%)]
[0150] After bottom grinding (A) and bottom etching (B), 5 inch
(127 mm) diameter GaN circular wafers of a 850 .mu.m thickness are
chamfered by a resin-bonding whetstone (C) with phenol resin as a
bonding material. This is not a rubber-bonding whetstone but a
resin-bonding whetstone, because the bonding material is a resin
(phenol). Unlike Sample 6, phenol resin is employed as a bonding
material. Phenol resin is not bubbled. The porosity is 0%. The
hardness cannot be measured by the method for rubber. Stationary
whetgranules are all diamond granules. No oxide granules are used.
The mesh number of diamond granules is #600 (average diameter 25
.mu.m), which is grosser than Samples 1-6 and equal to Sample 7.
The thickness d of the edge process-induced degradation layer
originating from the chamfering is d=13 .mu.m which is the thickest
among Samples 1-8. Though the mesh number is common to Sample 7,
non-bubbled phenol resin is harder than the CR rubber of Sample 7.
The mechanical function by harder phenol resin increases the edge
process-induced degradation layer thickness d over Sample 7. Edge
roughness of the postchamfered wafers is Ra7 .mu.m. The harder
phenol resin and grosser diamond granules raise the edge roughness.
The postchamfered GaN wafers are further planar-processed by top
surface grinding (D) and top surface polishing (E). The top surface
becomes mirror flat. The top process-induced degradation layer is
eliminated by top surface etching (F). The bow height U is 25 .mu.m
(R=80 m). The crack occurrence rate C is 12%, which is larger than
Samples 1-7. This fact means that no bubbled phenol resin brings
about a higher crack occurrence rate due to its hardness. The
device production yield Q by producing 2 mm square LED devices on
the wafer like the preceding samples is Q=50%. When 70% or more of
Q is required, Sample 8 is rejected.
[0151] Samples 1-8 teach that an allowable range of d is d=0.5
.mu.m-10 .mu.m from the standpoint of suppression of bow and crack
and enhancement of device yield. A more favorable range is d=1-3
.mu.m for the edge process-induced degradation layer thickness.
[Sample 9 (No Oxide Granules; Metal-Bond Whetstones (Cu--Sn); d=18
.mu.m; Q=20%)]
[0152] After bottom grinding (A) and bottom etching (B), 5 inch
(127 mm) diameter GaN circular wafers of a 850 .mu.m thickness are
chamfered by a metal-bond whetstone (C) with a Cu--Sn (copper/tin)
alloy as a bonding material. Since the bonding material is metal,
this is a metal bonding whetstone. The porosity is 0%. The hardness
is higher than the CR rubber. Stationary whetgranules are all
diamond granules. No oxide granules are used. The mesh number of
diamond granules is #600 (average diameter 25 .mu.m), which is
grosser than Samples 1-6 and equal to Samples 7-8. The thickness d
of the edge process-induced degradation layer originating from the
chamfering is d=18 .mu.m which is the thickest among Samples 1-9.
Though the diamond mesh number is common to Samples 7-8, the
bonding material Cu--Sn alloy of Sample 9 is harder than the rubber
or resin of Samples 7-8. The mechanical function of the bonding
metal increases the edge process-induced degradation layer,
thickness d over Samples 7-8. Edge roughness of the postchamfered
wafers is Ra8 .mu.m. Grosser diamond granules and harder metal
bonding raise the edge roughness. The postchamfered GaN wafers are
further planar-processed by top surface grinding (D) and top
surface polishing (E). The top surface becomes mirror flat. The top
process-induced degradation layer is eliminated by top surface
etching (F). The bow height U is 50 .mu.m (R=40 m). The crack
occurrence rate C is 26%, which is larger than Samples 1-8. The
device production yield Q by producing 2 mm square LED devices on
the wafer like the preceding samples is Q=20%, which is extremely
low. Sample 9 is rejected.
[0153] Samples 1-9 teach that a range d=2 .mu.m-6 .mu.m gives
minimum bow. Over d=6 .mu.m, the bow U increases with d. The crack
occurrence rate takes the minimum at d=2 .mu.m. Over d=2 .mu.m, the
crack occurrence rate C rises as d increases. The device production
yield Q is high within the range of d=1 .mu.m-3 .mu.m. Q is low
over d=3 .mu.m or under d=1 .mu.m.
[Sample 10 (Fe.sub.2O.sub.3 Granules; Metal-Bond Whetstones
(Cu--Sn); d=13 .mu.m; Q=52%)]
[0154] After bottom grinding. (A) and bottom etching (B), 5 inch
(127 mm) diameter GaN circular wafers of a 850 .mu.m thickness are
chamfered by a metal-bonding whetstone (C) with a Cu--Sn
(copper/tin) alloy as a bonding material. The porosity is 0%. The
hardness of the Cu--Sn alloy is higher than the CR rubber.
Stationary whetgranules are a mixture of diamond granules and
Fe.sub.2O.sub.3 (ferric oxide) granules. The mixture consists of 80
wt % diamond granules and 20 wt % Fe.sub.2O.sub.3 granules. The
mesh number of diamond granules is #3000 (average diameter 5
.mu.m), which is equal to Sample 2. The mixture ratio is identical
with Sample 2. What is different from Sample 2 is the bonding
material. The bonding material of Sample 2 is CR (chloroprene
rubber). The bonding material of Sample 10 is a Cu--Sn alloy.
Samples 2 and 10 clarify the difference between rubber-bonding and
metal-bonding whetstone chamfering. The thickness d of the edge
process-induced degradation layer originating from the chamfering
is d=13 .mu.m. The thickness d=13 .mu.m of Sample 10 is lower than
Sample 9 (d=18 .mu.m) chamfered by the same metal bonding
whetstone. The decrease of d results from the use of finer diamond
granules of Sample 10. The reason why d=13 .mu.m of Sample 10 is
larger than d=0.5 .mu.m of Sample 2 derives from the difference
between rubber bonding and metal bonding. Despite the same mesh
number and same mixture ratio, Samples 2 and 10 have different
thicknesses d=0.5 .mu.m and d=13 .mu.m. This result signifies that
rubber bonding whetstones abundant in elasticity are supreme
whetstones for chamfering. Edge roughness of the postchamfered
wafers is Ra1 .mu.m. Finer diamond granules and soft oxide granules
lowers the edge roughness. Metal bonding raises the edge roughness
over Sample 2. The postchamfered GaN wafers are further
planar-processed by top surface grinding (D) and top surface
polishing (E). The top surface becomes mirror flat. The top
process-induced degradation layer is eliminated by top surface
etching (F). The bow height U is 28 .mu.m (R=80 m). The crack
occurrence rate C is 14%, which is larger than Samples 1-8 but
smaller than Sample 9. Sample 10 teaches us that metal bonding
whetstones with hard bonding are unsuitable for chamfering, even if
fine diamond granules are used. The device production yield Q by
producing 2 mm square LED devices on the wafer like the preceding
samples is Q=52%. This is a low yield. 82% or more of device yields
Q require an optimum range of 70% or more of Q require a favorable
range of 0.5 .mu.m.ltoreq.d.ltoreq.10 .mu.m. Sample 10 is rejected
in the over-70% case. 50% or more of Q allow another range of 0.3
.mu.m.ltoreq.d.ltoreq.13 .mu.m.
[Sample 11 (No Oxide Granules; Electrodeposition Whetstones (Ni);
d=16 .mu.m; Q=35%)]
[0155] After bottom grinding (A) and bottom etching (B), 5 inch
(127 mm) diameter GaN circular wafers of a 850 .mu.m thickness are
chamfered by a diamond-electrodeposition whetstone (C) with nickel
(Ni) as a bonding material. This is an electrodeposition whetstone.
The porosity is 0%. The hardness of nickel (Ni) is higher than the
CR rubber. Stationary whetgranules are 100% diamond granules. The
whetstone includes no oxide granules. The mesh number of diamond
granules is #3000 (average diameter 5 .mu.m), which is the same as
Sample 10. What differs from Sample 10 is that Sample 11 contains
no oxide and a different bonding material. Diamond granules are
electrodeposited on a base disc. The thickness d of the edge
process-induced degradation layer originating from the chamfering
is d=16 .mu.m. The thickness d=16 .mu.m of Sample 11 is larger than
Sample 10 (d=13 .mu.m). The increase of d results from an increment
of rigidity by electrodeposition and the higher diamond granule
ratio. Edge roughness of the postchamfered wafers is Ra2 .mu.m.
Finer diamond granules lowers the edge roughness. But
electrodeposition raises edge roughness over Samples 3 and 10. The
postchamfered GaN wafers are further planar-processed by top
surface grinding (D) and top surface polishing (E). The top surface
becomes mirror flat. The top process-induced degradation layer is
eliminated by top surface etching (F). The bow height U is 43 .mu.m
(R=46 m). The crack occurrence rate C is 22%. Sample 11 reveals the
second highest bow U and crack occurrence rate C next to Sample 9.
This fact suggests a strong relation between d and U/C. Sample 11
teaches us the electrodeposition whetstones even furnished with
fine diamond granules are unsuitable for chamfering due to high
rigidity. The device production yield Q by producing 2 mm square
LED devices on the wafer like the preceding samples is Q=35%. This
is a very low yield. Sample 11 is rejected.
[0156] Samples 1-11 suggest that edge roughness has no direct
relation to d but depends upon the size and rate of diamond
granules. The edge roughness Ra increases with the diamond granule
size (smaller mesh number #). Inclusion of oxide granules to
whetstones gives edge roughness Ra lower than 100% diamond granule.
Samples 2, 3, and 11 suggest that rubber bonding gives the lowest
edge roughness and metal-bonding or electrodeposition is
indifferent to the edge roughness. There are not strong relations
among the edge roughness Ra, bow U, crack occurrence rate C and
device yield Q. What rules the bow, crack occurrence rate C and
device yield Q is the edge process-induced degradation layer
thickness d. The most significant parameter is d.
[0157] 70% or more device yields Q require d=0.5 .mu.m-10 .mu.m of
the edge process-induced degradation layer thickness (FIG. 16). 82%
or more device yields Q require d=1 .mu.m-3 .mu.m of the edge
process-induced degradation layer thickness. The range of d=3
.mu.m-6 .mu.m minimizes the bow U (FIG. 14). The value d=2 .mu.m
minimizes the crack occurrence rate C (FIG. 15). These values of d
coincide with the range of d which gives a high device yield. A
range d=0.5 .mu.m-10 .mu.m gives a high device yield Q, low crack
occurrence rate C and small bow U. A further desirable range is d=1
.mu.m-3 .mu.m.
[0158] A rise of the ratio and the size of diamond granules
increases the edge process-induced degradation layer thickness d.
Larger thicknesses d of the edge process-induced degradation layers
are induced by electrodeposition, metal-bonding, resin-bonding and
rubber-bonding whetstones in order. Soft rubber bonding whetstones
are pertinent to induce d=0.5 .mu.m-10 .mu.m of the edge
process-induced degradation layer thicknesses. Soft bubbled resin
bonding-whetstones are also available for making d=0.5 .mu.m-10
.mu.m (Sample 6). Favorable mesh numbers of diamond granules are
#3000-#600. Decrement of diamond ratio and increment of oxide ratio
are effective in further reducing d. Oxide granules produce soft
layers on surfaces of nitride by chemical reaction. Oxide granules
enables chamfering to alleviate a load and to suppress generation
of process-induced degradation layers. An edge roughness range
Ra0.07 .mu.m-Ra5 .mu.m corresponds to a thickness range d=0.5
.mu.m-10 .mu.m. A narrower edge roughness range Ra0.15 .mu.m-Ra1
.mu.m corresponds to another thickness range d=1 .mu.m-3 .mu.m.
[0159] In FIG. 14, 10 .mu.m or more thicknesses d of edge
process-induced degradation layers raise bow over 12 .mu.m. 0.3
.mu.m or less thicknesses d of edge process-induced degradation
layers enhance bow over 20 .mu.m. The fact means that the
thicknesses d=0.5 .mu.m-10 .mu.m can suppress bow under 12 .mu.m.
The thicknesses d=1 .mu.m-3 .mu.m can suppress bow within 7 .mu.m-5
.mu.m.
[0160] As shown in FIG. 15, 10 .mu.m or more thicknesses d of edge
process-induced degradation layers raise crack occurrence rates
over 5%. 0.3 .mu.m or less thicknesses d of edge process-induced
degradation layers push up crack occurrence rates over 9%. 5%-0% of
crack occurrence rates require d=0.5 .mu.m-10 .mu.m. 2%-0% of crack
occurrence rates are obtained by d=1 .mu.m-3 .mu.m.
[0161] In FIG. 16, 10 .mu.m or more thicknesses d of edge
process-induced degradation layers lower device production yields Q
below 70%. 0.3 .mu.m or less thicknesses d of edge process-induced
degradation layers lower Q below 52%. A pertinent range d=0.5
.mu.m-10 .mu.m can realize 70% or more device production yields Q.
A more favorable range d=1 .mu.m-3 .mu.m can raise device
production yields Q over 82%.
Embodiment 3
Embodiment 3
Samples 12-18: Table 2: FIGS. 17, 18, 19, 20, 28, 29
[0162] Samples 12-18 fix diamond granule size to #1000 and rubber
hardness to 50. Samples 12-18 examine edge oxygen concentrations,
chipping and device production yields by chamfering wafers by
varying oxides and oxide ratios. Preparation of wafers, order of
processings and method of device production of Embodiment 3 are
similar to Embodiment 2. Embodiment 3 examines how the edge oxygen
concentration rules wafer production and device production. Table 2
shows the results of experiments. Edge oxygen concentration is
represented by atomic percent (at %). Chipping means occurrence of
a slit or break of edges during planar-processings (grinding and
polishing). Device production yield Q is a ratio of the number of
allowable devices to the number of the whole devices which are
judged by producing 2 mm square LEDs on the wafer, packaging the
LEDs, supplying the LEDs with 4 A current, measuring light power
and examining the light power over or under 2 W.
TABLE-US-00002 TABLE 2 Sample Sample Sample Sample Sample Sample
Sample 12 13 14 15 16 17 18 Edge Bond- Kinds Rubber Rubber Rubber
Rubber Rubber Rubber Rubber Pro- ing Materials CR CR CR CR CR CR CR
cessing Rubber Hardness 50 50 50 50 50 50 50 Porosity (vol %) 0% 0%
0% 0% 0% 0% 0% Whet- Diamond(Mesh No.) 1000 1000 1000 1000 1000
1000 1000 granule Oxide -- Fe.sub.2O.sub.3 Cr.sub.2O.sub.3
Fe.sub.2O.sub.3 Cr.sub.2O.sub.3 MnO.sub.2 Fe.sub.2O.sub.3 Oxide
Amount(wt %) -- 20 20 30 30 40 60 Wafer Edge process-induced
degradation 5 2.5 2 2 1.5 1.5 1 Prop- layer thickness d (.mu.m)
erties Edge roughness Ra (.mu.m) 1.5 0.5 0.5 0.5 0.5 0.5 0.5 Oxygen
concentration O (at %) 1 3 5 10 15 20 25 Chipping occurrence rate p
(%) 25 5 3 2 3 5 15 Device Device production yield Q (%) 73 78 85
84 85 80 61
[0163] In chamfering, thicknesses d=0.5 .mu.m-10 .mu.m of edge
process-induced degradation layers are effective in suppressing
crack occurrence rates C and enhancing device production yields Q.
Thicknesses d=1 .mu.m-3 .mu.m of edge process-induced degradation
layers are more favorable. Samples 1-11 teach us that elastic
rubbers are optimum bonding materials and soft bubbled resins are
available. Samples 1-11 teach us suitable bonding materials. The
problem of what are the appropriate stationary granules is not
fully clarified. A whetstone of 100% diamond granules is possible
(Samples 3-7). Another whetstone of a mixture of diamond and oxide
granules is allowable (Sample 2). A middle mesh number between
Samples 5 and 6 is #1000. Embodiment 3 chamfers wafers by
whetstones having a mixture of #1000 diamond granules and oxide
granules. Chamfering is followed by planar-processings, i.e.,
top-grinding, top-polishing and top-etching. Embodiment 3 examines
chipping occurrence rates and final device yields. "Crack" means an
edge split or break occurring during chamfer. "Chipping" means an
edge split or break occurring in the following planar-processings.
Oxide granules can alleviate a load acting on edge in chamfering by
forming soft layers on surfaces of nitride with chemical reactions.
In addition to the advantage, the edge oxidized layers have a
function of suppressing chipping in the planar-processings.
Chamfering by a whetstone with oxide granules gives oxides to edges
of wafers. Existence of oxides gifts wafers with toughness. The
oxide-endowed toughness decreases chipping occurrence rates in the
planar-processings.
[0164] To confirm the function of oxides, Embodiment 3 chamfers
wafers by whetstones having a variety of kinds and amounts of oxide
granules. For separating the effect of oxides, the bonding material
is fixed to chloroprene rubber (CR). The hardness is fixed to 50.
The size of diamond granules is fixed to #1000. Variables are kinds
and amounts of oxides. FIG. 17 shows a relation between the ratio
of diamond granules (100 wt %-40 wt %) and oxide granules (0 wt
%-60 wt %) and the postchamfer edge oxygen concentration in Samples
12-18. FIG. 18 denotes a relation between the ratio of diamond
granules (100 wt %-40 wt %) and oxide granules (0 wt %-60 wt %) and
the thickness d (.mu.m) of the postchamfer edge process-induced
degradation layers in Samples 12-18. FIG. 19 shows a graph denoting
a relation between the ratio of diamond granules (100 wt %-40 wt %)
and oxide granules (0 wt %-60 wt %) and the chipping occurrence
rate p (%) in postchamfered planar-processings in Samples 12-18.
FIG. 20 is a graph showing a relation between the ratio of diamond
(100 wt %-40 wt %) and oxide (0 wt %-60 wt %) granules and the
device production yield Q(%) in Samples 12-18. FIG. 28 is a graph
showing a relation between the edge oxygen concentration (at %) and
the chipping occurrence rate p(%) in postchamfered
planar-processings in Samples 12-18. FIG. 29 is a graph showing a
relation between the edge oxygen concentration (at %) and the
device production yield Q (%).
[Sample 12 (No Oxide Granules; Rubber Whetstones; d=5 .mu.m; p=25%;
Q=73%)]
[0165] After bottom grinding (A) and bottom etching (B), 5 inch
(127 mm) diameter GaN circular wafers of a 850 .mu.m thickness are
chamfered by a chloroprene rubber (CR) bonding whetstone (C). The
porosity is 0%. Rubber hardness is 50. Stationary whetgranules are
100% diamond granules of #1000 (average diameter 15 .mu.m). The
whetstone includes no oxide granules. The thickness d of the edge
process-induced degradation layer originating from the chamfering
is d=5 .mu.m. Edge roughness of the postchamfered wafers is Ra1.5
.mu.m. The postchamfered GaN wafers are further planar-processed by
top surface grinding (D) and top surface polishing (E). The top
surface becomes mirror flat. The top process-induced degradation
layer is eliminated by top surface etching (F). Edge oxygen
concentration is O=1 at %. The edge oxygen does not derive from
whetstone granules but originates from oxidization in etching,
planar-processings, washing or natural oxidization in the
atmosphere. The chipping occurrence rate is p=25%. 25% of the
wafers are out of use. The device production yield Q by producing 2
mm square LED devices on the wafer like the preceding samples is
Q=73%. This is a high yield. Sample 12 is allowable. However such a
high chipping rate of p=25% is not desirable. Contrivance is
required for lowering the chipping rate.
[Sample 13 (20 wt % Fe.sub.2O.sub.3 Granules; Rubber Whetstones;
d=2.5 .mu.m; p=5%; Q=78%)]
[0166] After bottom grinding (A) and bottom etching (B), 5 inch
(127 mm) diameter GaN circular wafers of a 850 .mu.m thickness are
chamfered (C) by a rubber whetstone bonded with 80 wt % diamond
granules of #1000 (average diameter 15 .mu.m) and 20 wt %
Fe.sub.2O.sub.3 granules by chloroprene rubber (CR). The porosity
is 0%. Rubber hardness is 50. The thickness d of the edge
process-induced degradation layer originating from the chamfering
is d=2.5 .mu.m. Edge roughness of the postchamfered wafers is Ra0.5
.mu.m. The thickness d is reduced to half of Sample 12. Decrement
of diamond granules alleviates the mechanical function of grinding,
which results in reducing d. The edge roughness is smaller than
Sample 12. Oxide granules with a chemical function smooth the edge.
The postchamfered GaN wafers are further planar-processed by top
surface grinding (D) and top surface polishing (E). The top surface
becomes mirror flat. The top process-induced degradation layer is
eliminated by top surface etching (F). Edge oxygen concentration is
O=3 at %, which is three times as much as Sample 12 (O=1 at %). The
excess edge oxygen derives from the oxide whetstone granules. The
chipping occurrence rate is p=5%. The chipping occurrence rate is
reduced to one fifth of Sample 12 (p=25%). Existence of oxides at
the edge lessens the chipping occurrence rate p. The chemical
function of oxide granules decreases inner damage during chamfering
and suppresses wafers from chipping. Other samples with remaining
oxygen at edges have similar advantages of reducing inner stress
and prohibiting wafers from chipping. The device production yield Q
by producing 2 mm square LED devices on the wafer like the
preceding samples is Q=78%. This is a high yield. Sample 13 with a
low chipping occurrence rate and a high device yield is
approved.
[Sample 14 (20 wt % Cr.sub.2O.sub.3 Granules; Rubber Whetstones;
d=2 .mu.m; p=3%; Q=85%)]
[0167] After bottom grinding (A) and bottom etching (B), 5 inch
(127 mm) diameter GaN circular wafers of a 850 .mu.m thickness are
chamfered (C) by a rubber whetstone bonded with 80 wt % diamond
granules of #1000 (average diameter 15 .mu.m) and 20 wt %
Cr.sub.2O.sub.3 granules by chloroprene rubber (CR). The porosity
is 0%. Rubber hardness is 50. What is different from Sample 13 is
the use of stationary granules of Cr.sub.2O.sub.3 instead of
Fe.sub.2O.sub.3. The thickness d of the edge process-induced
degradation layer originating from the chamfering is d=2 .mu.m.
Edge roughness of the postchamfered wafers is Ra0.5 .mu.m. The
thickness d and edge roughness Ra are reduced in comparison with
Sample 12. Decrement of diamond granules alleviates the mechanical
function of grinding, which results in reducing d and Ra. Oxide
granules with a chemical function smooth the edge. The
postchamfered GaN wafers are further planar-processed by top
surface grinding (D) and top surface polishing (E). The top surface
becomes mirror flat. The top process-induced degradation layer is
eliminated by top surface etching (F). Edge oxygen concentration is
O=5 at %, which is five times as much as Sample 12 and 1.7 times as
much as Sample 13. Since chamfering uses oxide granule whetstones,
the edge oxygen concentration is high. The excess edge oxygen
derives from the oxide whetstone granules. Comparison of Samples 14
and 13 teaches us that Cr.sub.2O.sub.3 has an oxidization effect
stronger than Fe.sub.2O.sub.3. The chipping occurrence rate is
p=3%. The chipping occurrence rate is reduced to one eighth of
Sample 12 (p=25%). Existence of oxides at the edge lessens the
chipping occurrence rate p. Low chipping rates are an advantage of
the use of oxide granules. A high edge oxygen concentration more
than Sample 13 lowers the chipping occurrence rate of Sample 14.
The device production yield Q by producing 2 mm square LED devices
on the wafer like the preceding samples is Q=85%. This is a very
high yield. Sample 14 endowed with a low chipping occurrence rate
and a high device yield is most promising.
[Sample 15 (30 wt % Fe.sub.2O.sub.3 Granules; Rubber Whetstones;
d=2 .mu.m; p=2%; Q=84%)]
[0168] After bottom grinding (A) and bottom etching (B), 5 inch
(127 mm) diameter GaN circular wafers of a 850 .mu.m thickness are
chamfered (C) by a rubber whetstone bonded with 70 wt % diamond
granules of #1000 (average diameter 15 .mu.m) and 30 wt %
Fe.sub.2O.sub.3 granules by chloroprene rubber (CR). The porosity
is 0%. Rubber hardness is 50. The thickness d of the edge
process-induced degradation layer originating from the chamfering
is d=2 .mu.m. Edge roughness of the postchamfered wafers is Ra0.5
.mu.m. The thickness d (2 .mu.m) and edge roughness Ra (0.5 .mu.m)
are similar to Sample 14. Sample 15 uses Fe.sub.2O.sub.3 granules
like Sample 13. The oxide (Fe.sub.2O.sub.3) ratio of Sample 15 is
30% which is larger than Sample 13 by 10%. A smaller ratio of
diamond granules of Sample 15 weakens the mechanical function,
decreases the thickness d and lowers the edge roughness Ra. Oxide
granules with a chemical function smooth the edge. The
postchamfered GaN wafers are further planar-processed by top
surface grinding (D) and top surface polishing (E). The top surface
becomes mirror flat. The top process-induced degradation layer is
eliminated by top surface etching (F). Edge oxygen concentration is
O=10 at %. Sample 15 has an edge oxygen concentration higher than
Sample 13 which makes use of Fe.sub.2O.sub.3. This is because
Sample 15's whetstone has a Fe.sub.2O.sub.3 ratio higher than
Sample 13. The chipping occurrence rate is p=2%. The chipping
occurrence rate is one twelfth of Sample 12 which dispenses with
oxide granules. The chipping occurrence rate p of Sample 15 is
lower than Samples 13 and 14 which make use of oxide granules. The
lowest chipping occurrence rate of Sample 15 derives from the high
edge oxygen concentration. Samples 12-15 are evidences that what
prevents wafers from chipping is oxides included in edge of wafers.
Suppression of chipping is an advantage of making use of oxide
granules in whetstones. The device production yield Q by producing
2 mm square LED devices on the wafer like the preceding samples is
Q=84%. This is a very high yield. Sample 15 endowed with a low
chipping occurrence rate and a high device yield is most
desirable.
[Sample 16 (30 wt % Cr.sub.2O.sub.3 Granules; Rubber Whetstones;
d=1.5 .mu.m; p=3%; Q=85%)]
[0169] After bottom grinding (A) and bottom etching (B), 5 inch
(127 mm) diameter GaN circular wafers of a 850 .mu.m thickness are
chamfered (C) by a rubber whetstone bonded with 70 wt % diamond
granules of #1000 (average diameter 15 .mu.m) and 30 wt %
Cr.sub.2O.sub.3 granules by chloroprene rubber (CR). The porosity
is 0%. Rubber hardness is 50. The thickness d of the edge
process-induced degradation layer originating from the chamfering
is d=1.5 .mu.m. Edge roughness of the postchamfered wafers is Ra0.5
.mu.m. Sample 16 uses Cr.sub.2O.sub.3 granules like Sample 14. The
oxide (Cr.sub.2O.sub.3) ratio of Sample 16 is 30% which is larger
than Sample 14 by 10%. A smaller ratio of diamond granules of
Sample 16 weakens the mechanical function, decreases the thickness
d and lowers the edge roughness Ra. Sample 16 shows a low edge
roughness similar to Samples 14 and 15. The edge roughness depends
upon the diamond/oxide ratio but is irrelevant to the kind of
oxides. The postchamfered GaN wafers are further planar-processed
by top surface grinding (D) and top surface polishing (E). The top
surface becomes mirror flat. The top process-induced degradation
layer is eliminated by top surface etching (F). Edge oxygen
concentration is O=15 at %. Sample 16 (with 30% Cr.sub.2O.sub.3)
has an edge oxygen concentration higher than Sample 15 (with 30%
Fe.sub.2O.sub.3). This suggests that Cr.sub.2O.sub.3 has
oxidization effect stronger than Fe.sub.2O.sub.3. The chipping
occurrence rate is p=3%. The device production yield Q by producing
2 mm square LED devices on the wafer like the preceding samples is
Q=85%. This is a very high yield. Samples 14 and 16 have the same
chipping occurrence rate p and device yield Q, although the
Cr.sub.2O.sub.3 ratios of 20 wt % and 30 wt % are different. This
fact means that an optimum range of Cr.sub.2O.sub.3 rates is 20 wt
% to 30 wt %.
[Sample 17 (40 wt % MnO.sub.2 Granules; Rubber Whetstones; d=1.5
.mu.m; p=5%; Q=80%)]
[0170] After bottom grinding (A) and bottom etching (B), 5 inch
(127 mm) diameter GaN circular wafers of a 850 .mu.m thickness are
chamfered (C) by a rubber whetstone bonded with 60 wt % diamond
granules of #1000 (average diameter 15 .mu.m) and 40 wt % MnO.sub.2
granules by chloroprene rubber (CR). The porosity is 0%. Rubber
hardness is 50. The thickness d of the edge process-induced
degradation layer originating from the chamfering is d=1.5 .mu.m.
Edge roughness of the postchamfered wafers is Ra0.5 .mu.m. Sample
17 uses MnO.sub.2 (manganese dioxide) granules unlike Samples
13-16. The diamond ratio is smaller than Sample 12. The smaller
diamond ratio of Sample 17 weakens the mechanical function,
decreases the thickness d and lowers the edge roughness Ra. Sample
17 shows a low edge roughness (Ra0.5 .mu.m) similar to Samples
13-16. The postchamfered GaN wafers are further planar-processed by
top surface grinding (D) and top surface polishing (E). The top
surface becomes mirror flat. The top process-induced degradation
layer is eliminated by top surface etching (F). Edge oxygen
concentration is O=20 at %. Sample 17 which uses a whetstone
provided with a 40% high ratio of MnO.sub.2 has a high edge oxygen
concentration. The chipping occurrence rate is p=5%. The device
production yield Q by producing 2 mm square LED devices on the
wafer like the preceding samples is Q=80%. This is a high yield.
Samples 17 is allowable, since the chipping occurrence rate is low
and the device yield is high. Samples 12-17 teach us that MnO.sub.2
granules are also useful to the whetstone for chamfering.
[Sample 18 (60 wt % Fe.sub.2O.sub.3 Granules; Rubber Whetstones;
d=1 .mu.m; p=15%; Q=61%)]
[0171] After bottom grinding (A) and bottom etching (B), 5 inch
(127 mm) diameter GaN circular wafers of a 850 .mu.m thickness are
chamfered (C) by a rubber whetstone bonded with 40 wt % diamond
granules of #1000 (average diameter 15 .mu.m) and 60 wt %
Fe.sub.2O.sub.3 granules by chloroprene rubber (CR). The porosity
is 0%. Rubber hardness is 50. The thickness d of the edge
process-induced degradation layer originating from the chamfering
is d=1 .mu.m. Edge roughness of the postchamfered wafers is Ra0.5
.mu.m. Sample 17 uses Fe.sub.2O.sub.3 granules like Samples 13 and
15. In Sample 18, the oxide ratio is larger and the diamond ratio
is smaller than Samples 13 and 15. The smaller diamond ratio of
Sample 18 weakens the mechanical function and decreases the
thickness d. The postchamfered GaN wafers are further
planar-processed by top surface grinding (D) and top surface
polishing (E). The top surface becomes mirror flat. The top
process-induced degradation layer is eliminated by top surface
etching (F). Edge oxygen concentration is O=25 at %. Sample 18 has
a higher oxygen concentration than Samples 13 and 15. This is
because the Fe.sub.2O.sub.3 ratio is higher. The chipping
occurrence rate is p=15%. The device production yield Q by
producing 2 mm square LED devices on the wafer like the preceding
samples is Q=61%. This is not a high yield. Samples 18 is rejected
due to the high chipping occurrence rate (p=15%) and low device
yield (61%). Samples 12-18 teach us that 5% or less chipping
occurrence rates (p.ltoreq.5%) and 70% or more device yields
(Q.gtoreq.70%) require 3 at %-20 at % edge oxygen concentrations (3
at %.ltoreq.O.ltoreq.20 at %). 5% or less chipping rates
(p.ltoreq.5%) and 80% or more device yields (Q.gtoreq.80%) require
5 at %-20 at % edge oxygen concentrations (O=5 at %-20 at %).
Relations between the resultant edge oxygen concentration and the
oxide granule rate vary according to the kinds of oxides. 20 wt
%-40 wt % of oxide granules is effective in suppressing wafers from
chipping.
[0172] FIG. 17 indicates that the more the rate of oxide granules
bonded on the whetstones rises, the higher the edge oxygen
concentration increases. Comparison between Samples 13 and 14 and
comparison between Samples 15 and 16 teach us that Cr.sub.2O.sub.3
granules give wafer edges more residual oxygen than Fe.sub.2O.sub.3
even if Cr.sub.2O.sub.3 and Fe.sub.2O.sub.3 ratios are equal.
Chromium oxide is more reactive than ferric oxide. Chromium oxide
is harder than iron oxide. Chromium oxide has stronger influence on
chamfering than ferric oxide.
[0173] FIG. 18 shows that the smaller the oxide ratio decreases
(the larger the diamond rate rises), the thicker the edge
process-induced degradation layer grows. When the diamond ratio is
large, a strong mechanical function of diamond increases the edge
process-induced degradation layer thickness d from FIG. 18. It
turns out that oxides alleviate the diamond mechanical action.
[0174] FIG. 19 indicates that 20 wt %-40 wt % oxide rates can
suppress chipping occurrence rates to 5% or less. 100% diamond
granules (Sample 12) induce high rates of chipping. FIG. 28 teaches
us that 3 at %-20 at % edge oxygen concentrations can suppress
chipping occurrence rates below 5%. Sample 12, which chamfers
wafers with 100% diamond whetstones and leaves oxygen at edges at a
low rate below 1 at %, is suffering from high frequency of
chipping.
[0175] FIG. 20 shows that 20 wt %-40 wt % oxide rates can sustain
device production yields above 78%. FIG. 29 teaches us that 3 at
%-20 at % edge oxygen concentrations can raise device production
yields up to 78% or more.
Embodiment 4
Embodiment 4
Samples 19-26; Table 3; FIGS. 21, 22, 23, 30, 31
[0176] Embodiments 1-3 clarify that GaN wafer chamfering requires
d=0.5 .mu.m-10 .mu.m, in particular, d=1 .mu.m-3 .mu.m thicknesses
of edge process-induced degradation layers for the sake of
suppressing cracks and raising device yields, and rubber bonding
whetstones are favorable. 3 at %-20 at % edge oxygen concentrations
advantageously lower chipping frequencies and enhance device
production yields. Mixing of oxide granules into whetstone
stationary granules is effective in lifting edge oxygen
concentrations.
TABLE-US-00003 TABLE 3 Sample Sample Sample Sample Sample Sample
Sample Sample 19 20 21 22 23 24 25 26 Edge Bond- Kinds Rubber
Rubber Rubber Rubber Rubber Rubber Metal Electro- Pro- ing Deposi-
cessing tion Materials CR CR CR CR CR CR Fe Ni Rubber Hardness 55
55 55 55 55 55 -- -- Porosity (vol %) 0% 0% 0% 0% 0% 0% 0% 0% Whet-
Diamond(Mesh No.) 600 600 600 600 600 600 1500 1500 granule Oxide
-- Fe.sub.2O.sub.3 Cr.sub.2O.sub.3 ZnO CuO CuO -- -- Oxide
Amount(wt %) -- 20 20 20 20 50 -- -- Wafer Edge process-induced
degradation 9 7 7 7 7 5 12 14 Prop- layer thickness d (.mu.m)
erties Edge roughness Ra (.mu.m) 4 2 2 2 2 2 4 4 Metal
concentration m (at %) 0 0.1 0.2 3 5 8 10 12 Chipping occurrence
rate p (%) 22 8 4 7 15 15 25 30 Device Device production yield Q
(%) 90 92 90 88 85 65 35 22
[0177] Samples 19-26 investigate how much resultant metal
impurities can remain at wafer edges by chamfering wafers with
whetstones having a variety of ratios and kinds of oxide granules.
In Table 3, metal amount (at %) means the metal concentration at
edges after chamfering sample wafers.
[0178] FIG. 21 shows a relation between edge process-induced
degradation layer thicknesses d (.mu.m) and edge metal
concentrations (at %) in Samples 19-26. A vertical dotted line d=10
.mu.m shows an upper critical thickness of the present
invention.
[0179] FIG. 22 shows a relation between process-induced degradation
layer thicknesses d (.mu.m) and chipping occurrence rates p (%) in
Samples 19-26. A vertical dotted line d=10 .mu.m shows an upper
critical thickness of the present invention.
[0180] FIG. 23 shows a relation between process-induced degradation
layer thicknesses d (.mu.m) and device production yields Q (%) in
Samples 19-26. A vertical dotted line d=10 .mu.m shows an upper
critical thickness of the present invention. FIG. 30 shows a
relation between edge metal concentrations m (at %) and chipping
occurrence rates p(%) in Samples 19-26. FIG. 31 shows a relation
between edge metal concentrations m (at %) and device production
yields Q(%) in Samples 19-26.
[Sample 19 (No Oxide; m=at 0%; d=9 .mu.m; p=22%; Q=90%)]
[0181] After bottom grinding (A) and bottom etching (B), 5 inch
(127 mm) diameter GaN circular wafers of a 850 .mu.m thickness are
chamfered (C) by a rubber whetstone bonded with 100 wt % diamond
granules of #600 (average diameter 25 .mu.m) by chloroprene rubber
(CR). The porosity is 0%. Rubber hardness is 55. The thickness d of
the edge process-induced degradation layer originating from the
chamfering is d=9 .mu.m. Edge roughness of the postchamfered wafers
is Ra4 .mu.m. The postchamfered GaN wafers are further
planar-processed by top surface grinding (D) and top surface
polishing (E). The top surface becomes mirror flat. The top
process-induced degradation layer is eliminated by top surface
etching (F). Edge metal concentration is m=0 at %. Sample 19 uses
whetstones without metal oxide. Metal impurities do not invade
edges of Sample 19. The chipping occurrence rate is p=22%. 22%
wafers are wasted. The device production yield Q by producing 2 mm
square LED devices on the wafer like the preceding samples is
Q=90%. This is a high yield. Despite the high yield, Samples 19 is
rejected due to the high chipping occurrence rate (p=22%) in the
preceding planar-processings. The large diamond granules enhance
mechanical shocks to wafers and cause inner damage in wafers.
[Sample 20 (20 wt % Fe.sub.2O.sub.3; m=0.1 at %; d=7 .mu.m; p=8%;
Q=92%)]
[0182] After bottom grinding (A) and bottom etching (B), 5 inch
(127 mm) diameter GaN circular wafers of a 850 .mu.m thickness are
chamfered (C) by a rubber whetstone bonded with 20 wt %
Fe.sub.2O.sub.3 granules and 80 wt % diamond granules of #600
(average diameter 25 .mu.m) by chloroprene rubber (CR). The
porosity is 0%. Rubber hardness is 55. The thickness d of the edge
process-induced degradation layer originating from the chamfering
is d=7 .mu.m. Edge roughness of the postchamfered wafers is Ra2
.mu.m. The postchamfered GaN wafers are further planar-processed by
top surface grinding (D) and top surface polishing (E). The top
surface becomes mirror flat. The top process-induced degradation
layer is eliminated by top surface etching (F). Edge metal
concentration is m=0.1 at %. Metal impurities invade edges from
oxide granules of Sample 20. The chipping occurrence rate is p=8%.
8% is an allowable value for p. The device production yield Q by
producing 2 mm square LED devices on the wafer like the preceding
samples is Q=92%. Sample 20 is allowable due to the low chipping
occurrence rate p and the high device yields Q. Since Sample 20
uses ferric oxide granules, iron atoms remain at 0.1 at % at edges.
Metal atoms and oxygen atoms lower d and p. Metal atoms and oxygen
atoms reinforce GaN crystal structures. While chamfering, metal and
oxygen alleviate inner shocks and inner damage.
[Sample 21 (20 wt % Cr.sub.2O.sub.3; m=0.2 at %; d=7 .mu.m; p=4%;
Q=90%)]
[0183] After bottom grinding (A) and bottom etching (B), 5 inch
(127 mm) diameter GaN circular wafers of a 850 .mu.m thickness are
chamfered (C) by a rubber whetstone bonded with 20 wt %
Cr.sub.2O.sub.3 granules and 80 wt % diamond granules of #600
(average diameter 25 .mu.m) by chloroprene rubber (CR). The
porosity is 0%. Rubber hardness is 55. The thickness d of the edge
process-induced degradation layer originating from the chamfering
is d=7 .mu.m. Edge roughness of the postchamfered wafers is Ra2
.mu.m. The postchamfered GaN wafers are further planar-processed by
top surface grinding (D) and top surface polishing (E). The top
surface becomes mirror flat. The top process-induced degradation
layer is eliminated by top surface etching (F). Edge metal
concentration is m=0.2 at %. The chipping occurrence rate is p=4%.
4% is an allowable value for p. The device production yield Q by
producing 2 mm square LED devices on the wafer like the preceding
samples is Q=90%. Sample 21 is allowable due to the low chipping
occurrence rate p and the high device yields Q. Since Sample 21
uses chromic oxide granules, chromium (Cr) atoms remain at 0.2 at %
at edges. Metal atoms and oxygen atoms lower d and p. Metal atoms
and oxygen atoms reinforce GaN crystal structures. While
chamfering, metal and oxygen alleviate inner shocks and inner
damage.
[Sample 22 (20 wt % ZnO; m=3 at %; d=7 .mu.m; p=7%; Q=88%)]
[0184] After bottom grinding (A) and bottom etching (B), 5 inch
(127 mm) diameter GaN circular wafers of a 850 .mu.m thickness are
chamfered (C) by a rubber whetstone bonded with 20 wt % ZnO
granules and 80 wt % diamond granules of #600 (average diameter 25
.mu.m) by chloroprene rubber (CR). The porosity is 0%. Rubber
hardness is 55. The thickness d of the edge process-induced
degradation layer originating from the chamfering is d=7 .mu.m.
Edge roughness of the postchamfered wafers is Ra2 .mu.m. The
postchamfered GaN wafers are further planar-processed by top
surface grinding (D) and top surface polishing (E). The top surface
becomes mirror flat. The top process-induced degradation layer is
eliminated by top surface etching (F). Edge metal concentration is
m=3 at %. The chipping occurrence rate is p=7%. 7% is an allowable
value for p. The device production yield Q by producing 2 mm square
LED devices on the wafer like the preceding samples is Q=88%.
Sample 22 is allowable due to the low chipping occurrence rate p
and the high device yield Q. Since Sample 22 uses zinc oxide
granules, zinc (Zn) atoms remain at 3 at % at edges. Metal atoms
and oxygen atoms lower d and p. Metal atoms and oxygen atoms
reinforce GaN crystal structures. While chamfering, metal and
oxygen alleviate inner shocks and inner damage.
[Sample 23 (20 wt % CuO; m=5 at %; d=7 .mu.m; p=15%; Q=85%)]
[0185] After bottom grinding (A) and bottom etching (B), 5 inch
(127 mm) diameter GaN circular wafers of a 850 .mu.m thickness are
chamfered (C) by a rubber whetstone bonded with 20 wt % CuO
granules and 80 wt % diamond granules of #600 (average diameter 25
.mu.m) by chloroprene rubber (CR). The porosity is 0%. Rubber
hardness is 55. The thickness d of the edge process-induced
degradation layer originating from the chamfering is d=7 .mu.m.
Edge roughness of the postchamfered wafers is Ra2 .mu.m. The
postchamfered GaN wafers are further planar-processed by top
surface grinding (D) and top surface polishing (E). The top surface
becomes mirror flat. The top process-induced degradation layer is
eliminated by top surface etching (F). Edge metal concentration is
m=5 at %. The chipping occurrence rate is p=15%. The device
production yield Q by producing 2 mm square LED devices on the
wafer like the preceding samples is Q=85%. Although p=15% is not a
low chipping rate, Sample 22 is allowable due to the high device
yield Q. Since Sample 23 uses copper oxide granules, copper (Cu)
atoms remain at 5 at % at edges. Metal atoms and oxygen atoms lower
d and p. Metal atoms and oxygen atoms reinforce GaN crystal
structures. While chamfering, metal and oxygen alleviate inner
shocks and inner damage. Samples 20-23 have common 20 wt % oxide
rates. Amounts of residual metals at edges are different. Edge
resultant metals are ranked in order of amounts, i.e., Cu (Sat %),
Zn (3 at %), Cr (0.2 at %) and Fe(0.1 at %). Edge concentrations of
residual metals depend upon chemical reactivity, granule hardness,
and washing removability of oxide granules.
[Sample 24 (50 wt % CuO; m=8 at %; d=5 .mu.m; p=15%; Q=65%)]
[0186] After bottom grinding (A) and bottom etching (B), 5 inch
(127 mm) diameter GaN circular wafers of a 850 .mu.m thickness are
chamfered (C) by a rubber whetstone bonded with 50 wt % CuO
granules and 50 wt % diamond granules of #600 (average diameter 25
.mu.m) by chloroprene rubber (CR). The porosity is 0%. Rubber
hardness is 55. The thickness d of the edge process-induced
degradation layer originating from the chamfering is d=5 .mu.m.
Edge roughness of the postchamfered wafers is Ra2 .mu.m. The
postchamfered GaN wafers are further planar-processed by top
surface grinding (D) and top surface polishing (E). The top surface
becomes mirror flat. The top process-induced degradation layer is
eliminated by top surface etching (F). Edge metal concentration is
m=8 at %. The chipping occurrence rate is p=15%. The device
production yield Q by producing 2 mm square LED devices on the
wafer like the preceding samples is Q=65%. Since the chipping rate
p is high and the device yield Q is low, Sample 23 is rejected.
Since Sample 24 uses whetstones of 50 wt % CuO granules, large
impurity atoms lower the device yield Q. An m=0.2 at % edge metal
concentration minimizes the chipping occurrence rates p (Sample
21). An m=0.1 at % edge metal concentration maximizes the device
production yield Q (Sample 20).
[0187] Resultant metals are useful. But too large metal remains are
not desirable. m=0.1 at %-5 at % is a preferable range of edge
metal concentrations. More favorable range is m=0.1 at %-3 at
%.
[0188] Samples 19-23 declare that oxide including whetstone
chamfering can suppress chipping occurrence rates p lower than pure
diamond whetstone chamfering.
[0189] Metal impurities residing at edges prevent wafers from
chipping in Samples 20-23. Device production yields Q are
irrelevant to the edge metal remains in Samples 19-23.
[0190] Sample 19, which uses 100% diamond granule whetstones, has
d=9 .mu.m. Samples 20-23, which commonly use 80 wt % diamond
granule whetstones, have d=7 .mu.m. The edge process-induced
degradation layer thickness d is determined by the diamond granule
rate in whetstones in view of Samples 19-24.
[Sample 25 (No oxide; metal bond (Fe); m=10 at %; d=12 .mu.m;
p=25%; Q=35%)]
[0191] After bottom grinding (A) and bottom etching (B), 5 inch
(127 mm) diameter GaN circular wafers of a 850 .mu.m thickness are
chamfered (C) by a metal bonding whetstone bonded with 100 wt %
diamond granules of #1500 (average diameter 8 .mu.m) by iron (Fe).
The thickness d of the edge process-induced degradation layer
originating from the chamfering is d=12 .mu.m. Edge roughness of
the postchamfered wafers is Ra4 .mu.m. What increases the thickness
d and edge roughness Ra is the high rate of diamond and the bonding
metal. The postchamfered GaN wafers are further planar-processed by
top surface grinding (D) and top surface polishing (E). The top
surface becomes mirror flat. The top process-induced degradation
layer is eliminated by top surface etching (F). Edge metal
concentration is m=10 at %. Sample 25 uses whetstones without metal
oxide. But the bonding material is iron (Fe). Iron atoms invade
edges. 10 at % iron (Fe) atoms remain at edges. The chipping
occurrence rate is p=25%. The device production yield Q by
producing 2 mm square LED devices on the wafer like the preceding
samples is Q=35%. Sample 25 is rejected due to the high chipping
occurrence rate p and low device yield Q. The high chipping
occurrence rate and low device yield originate from the excess
thick process-induced degradation layers and excess high metal
concentration (10 at %). The reason is that Sample 25 uses
metal-bonded whetstones. Sample 25 suggests incompetency of
metal-bonded whetstones.
[Sample 26 (No Oxide; Electrodeposition (Ni); m=12 at %; d=14
.mu.m; p=30%; Q=22%)]
[0192] After bottom grinding (A) and bottom etching (B), 5 inch
(127 mm) diameter GaN circular wafers of a 850 .mu.m thickness are
chamfered (C) by an electrodeposition whetstone electrodeposited by
nickel (Ni) with 100 wt % diamond granules of #1500 (average
diameter 8 .mu.m). The thickness d of the edge process-induced
degradation layer originating from the chamfering is d=14 .mu.m.
Edge roughness of the postchamfered wafers is Ra4 .mu.m. What
raises the thickness d and edge roughness Ra is the high rate of
diamond and the rigid nickel electrodeposition. The postchamfered
GaN wafers are further planar-processed by top surface grinding (D)
and top surface polishing (E). The top surface becomes mirror flat.
The top process-induced degradation layer is eliminated by top
surface etching (F). Edge metal concentration is m=12 at %. Sample
26 uses whetstones without metal oxide. But diamond granules are
electrodeposited by nickel. A part of nickel is ground away. Nickel
atoms invade edges of wafers. 12 at % nickel (Ni) atoms remain at
edges. The chipping occurrence rate is p=30%. The device production
yield Q by producing 2 mm square LED devices on the wafer like the
preceding samples is Q=22%. Sample 26 is rejected due to the high
chipping occurrence rate p and low device yield Q. The high
chipping occurrence rate and low device yield result from the
excess thick process-induced degradation layers (d=14 .mu.m) and
excess high metal concentration (12 at %). The reason is that
Sample 26 uses electrodeposition whetstones. Sample 26 suggests
incompetency of electrodeposition whetstones.
[0193] Samples 19-26 declare that 15% or less chipping occurrence
rates p and 80% or more device yields Q require 10 .mu.m or less
thick edge process-induced degradation layers and 0.1 at %-5 at %
edge metal concentrations m. A narrower range of m=0.1 at %-3 at %
enables wafers to obtain 8% or less chipping occurrence rates p and
88% or more device yields Q.
[0194] FIG. 24 shows a relation between edge process-induced
degradation layer thicknesses d (.mu.m) and edge roughnesses Ra
(.mu.m) for all Samples 1-26.
[0195] Increment of edge process-induced degradation layer
thicknesses d increases edge roughness. Gross diamond granules and
low oxide rates reinforce mechanical action and enlarge edge
process-induced degradation layer thicknesses d and edge
roughnesses Ra. Thus d and Ra rise or set in rough proportion in a
main series. However, there is another series of samples--Samples
10 and 11--, which deviate from the main series in which d is in
proportion to Ra. Metal-bonding or metal-electrodeposition
whetstones have highly rigid bonding materials. High rigidity of
metals raises edge process-induced degradation layer thicknesses d,
maintaining small edge roughnesses.
[0196] Metal-bonded whetstones and electrodeposition whetstones are
incompetent to chamfering GaN wafers. Rubber-bonded or bubbled
resin-bonded whetstones are competent. Edge process-induced
degradation layer thicknesses d are more suitable than edge
roughnesses Ra for estimating the performance of GaN wafer
chamfering. FIG. 30 indicates a relation between edge metal
concentrations m (at %) and chipping occurrence rates p (%) in
Samples 19-26. 0.1 at %-5 at % edge metal concentrations enable GaN
wafers to lower chipping occurrence rates to 15% or less. FIG. 31
exhibits a relation between edge metal concentrations m (at %) and
device production yields Q (%) in Samples 19-26. 0 at %-5 at % edge
metal concentrations can maintain 85% or more device yields Q.
* * * * *