U.S. patent application number 16/851455 was filed with the patent office on 2021-10-21 for analog computing architecture for four terminal memory devices.
The applicant listed for this patent is International Business Machines Corporation. Invention is credited to Kevin W. Brew, Lawrence A. Clevenger, Timothy Mathew Philip.
Application Number | 20210327502 16/851455 |
Document ID | / |
Family ID | 1000005880483 |
Filed Date | 2021-10-21 |
United States Patent
Application |
20210327502 |
Kind Code |
A1 |
Philip; Timothy Mathew ; et
al. |
October 21, 2021 |
ANALOG COMPUTING ARCHITECTURE FOR FOUR TERMINAL MEMORY DEVICES
Abstract
A multiterminal non-volatile memory cross-bar array system
includes a set of conductive row rails, a set of conductive column
rails configured to form a plurality of crosspoints at
intersections between the conductive rails and the conductive
column rails and a resistive processing unit at each of the
crosspoints each representing a neuron in a neural network. At
least one given conductive row rail includes first and second row
lines is in contact with a given resistive processing unit. At
least one given conductive column rail including first and second
column lines is in contact with the given resistive processing
unit.
Inventors: |
Philip; Timothy Mathew;
(Albany, NY) ; Brew; Kevin W.; (Albany, NY)
; Clevenger; Lawrence A.; (Saratoga Springs, NY) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
International Business Machines Corporation |
Armonk |
NY |
US |
|
|
Family ID: |
1000005880483 |
Appl. No.: |
16/851455 |
Filed: |
April 17, 2020 |
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
G11C 13/0004 20130101;
G11C 13/004 20130101; G06N 3/063 20130101; G11C 13/0026 20130101;
G11C 2213/79 20130101; G11C 13/0028 20130101; G11C 13/003 20130101;
G11C 13/0069 20130101; G11C 11/54 20130101 |
International
Class: |
G11C 11/54 20060101
G11C011/54; G11C 13/00 20060101 G11C013/00; G06N 3/063 20060101
G06N003/063 |
Claims
1. A non-volatile memory cross-bar array system, which comprises: a
set of conductive row rails; a set of conductive column rails
configured to form a plurality of crosspoints at intersections
between the conductive rails and the conductive column rails; a
resistive processing unit at each of the crosspoints, the resistive
processing units each representing a neuron in a neural network; at
least one given conductive row rail including first and second row
lines in contact with a given resistive processing unit; and at
least one given conductive column rail including first and second
column lines in contact with the given resistive processing
unit.
2. The cross-bar array system of claim 1 wherein: the first row
line is a dedicated row write line and the second row line is a
dedicated row read line; and the first column line is a dedicated
column write line and the second column line is a dedicated column
read line.
3. The cross-bar array system of claim 2 including a first
transistor coupled to each of the row write line and the column
write line, the first transistors configured to control flow of
current through the row write line and the column write line.
4. The cross-bar array system of claim 3 including a second
transistor coupled to each of the row read line and the column read
line, the second transistors configured to control flow of current
through the row read line and the column read line.
5. The cross-bar array system of claim 2 including: a pMOS
transistor coupled to one of the row write line and the row read
line of the at least one given row rail; an nMOS transistor coupled
to the other of the row write line and the row read line; an nMOS
transistor coupled to the one of the column write line and the
column read line of the at least one given column rail; and a pMOS
transistor coupled to the other of the column write line and the
column read line.
6. The cross-bar array system of claim 5 wherein: one of the row
write line and the column write line is coupled to a pMOS
transistor and the other of the row write line and the column write
line is coupled to an nMOS transistor; and one of the row read line
and the column read line is coupled to a pMOS transistor and the
other of the row read line and the column read line is coupled to
an nMOS transistor;
7. The cross-bar array system of claim 2 wherein: the row write
line and the column write line comprise a first material having a
first level of resistivity; and the row read line and the column
read line comprise a second material having a second level of
resistivity less than the first level of resistivity.
8. The cross-bar array system of claim 7 including a transistor
coupled to each of the of the row read line and the column read
line.
9. The cross-bar array system of claim 2 wherein: each of the
conductive row rails include a row write line and a row read line
in contact with individual ones of the resistive processing units
and wherein each of the column rails include a column write line
and a column read line in contact with individual ones of the
resistive processing units.
10. The cross-bar array system of claim 1 wherein the resistive
processing unit cells comprise random access memory devices.
11. A non-volatile memory cross-bar array system, comprises: a set
of conductive row rails, each conductive rail including a pair of
row lines; a set of conductive column rails configured to form a
plurality of crosspoints at intersections between the conductive
row rails and the conductive column rails, each column rail
including a pair of column lines; and a matrix of resistive
processing unit cells, individual resistive processing unit cells
being electrically connected at respective crosspoints to
electrically couple with respective row rails and column rails;
wherein a given resistive processing unit cell comprises a four
terminal memory device.
12. The cross-bar array system of claim 11 wherein: at least one
given conductive row rail includes a row write line and a row read
line in contact with the given resistive processing unit cell; and
at least one given conductive column rail includes a column write
line and a column read line in contact with the given resistive
processing unit cell.
13. The cross-bar array system of claim 12 including: one or more
transistors to selectively control flow of current through the row
write line and the column write line; and one or more transistors
to selectively control flow of current through the row read line
and the column read line.
14. The cross-bar array system of claim 13 including: a pMOS
transistor coupled to one of the row write line and the row read
line of the at least one given row rail; an nMOS transistor coupled
to the other of the row write line and the row read line; an nMOS
transistor coupled to the one of the column write line and the
column read line of the at least one given column rail; and a pMOS
transistor coupled to the other of the column write line and the
column read line.
15. The cross-bar array system of claim 14 wherein: one of the row
write line and the column write line is coupled to a pMOS
transistor and the other of the row write line and the column write
line is coupled to an nMOS transistor; and one of the row read line
and the column read line is coupled to a pMOS transistor and the
other of the row write line and the column write line is coupled to
an nMOS transistor;
16. The cross-bar array system of claim 12 wherein: the row write
line and the column write line comprise a first material having a
first level of resistivity; and the row read line and the column
read line comprise a second material having a second level of
resistivity less than the first level.
17. The cross-bar array system of claim 16 including a transistor
coupled to each of the row read line and the column read line to
selectively prevent current from flowing through the row read line
and the column read line.
18. A non-volatile memory cross-bar array system, which comprises:
a set of conductive row rails; a set of conductive column rails
configured to form a plurality of crosspoints at intersections
between the conductive rails and the conductive column rails; a
resistive processing unit at each of the crosspoints, the resistive
processing units each representing a neuron in a neural network; at
least one given conductive row rail including a row write line and
a row read line in contact with a given resistive processing unit;
at least one given conductive column rail including a column write
line and a column read line in contact with the given resistive
processing unit; wherein the row write line and the column write
line comprise a first material having a first level of resistivity;
and wherein the row read line and the column read line comprise a
second material having a second level of resistivity less than the
first level.
19. The cross-bar array system of claim 18 including a transistor
coupled to each of the of the row read line and the column read
line.
20. The cross-bar array system of claim 19 wherein the transistor
is an nMOS transistor.
Description
BACKGROUND
[0001] The present invention relates generally to computing
architectures for an artificial neural network (ANN), and, in
particular, to a multiterminal cross-bar array system for
neuromorphic or artificial intelligence computing in conjunction
with an ANN.
[0002] Artificial neural networks (ANNs) can be formed from
cross-bar arrays of resistive processing units (RPUs) that provide
local data storage and local data processing. Cross-bar arrays are
high density, low cost circuit architectures used to form a variety
of electronic circuits and devices, including ANN architectures,
neuromorphic microchips and ultra-high density nonvolatile memory
devices. A basic cross-bar array configuration includes a set of
conductive row wires and a set of conductive column wires formed to
intersect the set of conductive row wires. The intersections
between the two sets of wires are separated by cross-point memory
devices. Trainable resistive memory cross-point devices are
referred to as resistive processing unit (RPU) cells. RPU cells are
considered as a promising technology for electronic synapse devices
or memristors for neuromorphic computing as well as high-density
and high-speed non-volatile memory applications. For example, in
neuromorphic computing applications, an RPU cell can be used as a
connection (synapse) between a pre-neuron and post-neuron.,
representing the connection weight in the form of device
resistance. Multiple pre-neurons and post-neurons can be connected
through a cross-bar array of the RPU cells, which naturally
expresses a fully-connected neural network.
SUMMARY
[0003] In an illustrative embodiment, a non-volatile memory
cross-bar array system comprises a set of conductive row rails, a
set of conductive column rails configured to form a plurality of
crosspoints at intersections between the conductive rails and the
conductive column rails, and a resistive processing unit at each of
the crosspoints each representing a neuron in a neural network. At
least one given conductive row rail includes first and second row
lines in contact with a given resistive processing unit. At least
one given conductive column rail includes first and second column
lines in contact with the given resistive processing unit.
[0004] In another illustrative embodiment, a non-volatile memory
cross-bar array system comprises a set of conductive row rails and
a set of conductive column rails configured to form a plurality of
crosspoints at intersections between the conductive rails and the
conductive column rails. Each conductive rail includes a pair of
row lines and each column rail includes a pair of column lines. The
cross-bar array system further includes a matrix of resistive
processing unit cells. Individual resistive processing unit cells
are disposed at the crosspoints to electrically couple with
respective row rails and column rails. A given resistive processing
unit cell comprises a four terminal memory device.
[0005] In another illustrative embodiment, a non-volatile memory
cross-bar array system comprises a set of conductive row rails, a
set of conductive column rails configured to form a plurality of
crosspoints at intersections between the conductive rails and the
conductive column rails, and a resistive processing unit at each of
the crosspoints representing a neuron in a neural network. At least
one given conductive row rail includes a row write line and a row
read line in contact with a given resistive processing unit. At
least one given conductive column rail includes a column write line
and a column read line in contact with the given resistive
processing unit. The row write line and the column write line
comprise a first material having a first level of resistivity. The
row read line and the column read line comprise a second material
having a second level of resistivity less than the first level.
BRIEF DESCRIPTION OF THE DRAWINGS
[0006] FIG. 1 is a schematic diagram illustrating an example
neuromorphic device architecture according to an illustrative
embodiment of the invention.
[0007] FIG. 2 is a schematic view of an illustrative embodiment of
a cross-bar array system for use with the neuromorphic device
architecture of FIG. 1.
[0008] FIG. 3 is a schematic view of another illustrative
embodiment of a cross-bar array system for use with the
neuromorphic device architecture of FIG. 1.
[0009] FIG. 4 is a schematic view of an illustrative embodiment of
a cross-bar array system for use with the neuromorphic device
architecture of FIG. 1.
DETAILED DESCRIPTION
[0010] The present invention relates in general to a semiconductor
device. More specifically, the present invention relates to an
analog computing architecture having application for use in
artificial neural networks (ANNs). The computing architecture is
formed from cross-bar arrays of multi-terminal resistive processing
units (RPUs) that provide local data storage and local data
processing to facilitate implementation of artificial intelligence
algorithms for matrix inversion, matrix decomposition and the
like.
[0011] In the following discussion, the term "illustrative" is used
herein to mean "serving as an example, instance or illustration."
Any embodiment or design described herein is intended to be
"illustrative" and is not necessarily to be construed as preferred
or advantageous over other embodiments or designs. The term
"connection" or "connected" can include both an indirect
"connection" and a direct "connection." The terms "on" or "onto"
with respect to placement of components relative to the
semiconductor structure or stack are not to be interpreted as
requiring direct contact of the components for it is possible one
or more intermediate components, layers or coatings may be
positioned between the select components unless otherwise
specified. More specifically, positional relationships, unless
specified otherwise, can be direct or indirect, and the present
invention is not intended to be limiting in this respect.
[0012] ANNs are often embodied as "neuromorphic" systems of
interconnected processor elements that act as simulated "neurons"
and exchange "messages" between each other in the form of
electronic signals. Similar to the "plasticity" of synaptic
neurotransmitter connections that carry messages between biological
neurons, the connections in ANNs that carry electronic messages
between simulated neurons are provided with numeric weights that
correspond to the strength or weakness of a given connection. The
weights can be adjusted and tuned based on experience, making ANNs
adaptive to inputs and capable of learning. For example, an ANN for
handwriting recognition is defined by a set of input neurons which
can be activated by the pixels of an input image. After being
weighted and transformed by a function determined by the network's
designer, the activations of these input neurons are then passed to
other downstream neurons, which are often referred to as "hidden"
neurons. This process is repeated until an output neuron is
activated. The activated output neuron determines which character
was read.
[0013] Resistive processing units (RPU) cells, in effect, function
as the ANN's weighted connections between neurons. Nanoscale
two-terminal devices, for example, memristors having "ideal"
conduction state switching characteristics, are often used as the
RPU cells coupled to a cross-bar array to emulate synaptic
plasticity with high energy efficiency. The conduction state (e.g.,
resistance) of the ideal memristor material can be altered by
controlling the voltages applied between individual wires of the
row and column wires. Digital data can be stored by alteration of
the conduction state of the memristor material at the intersection
to achieve a high conduction state or a low conduction state. The
memristor material can also be programmed to maintain two or more
distinct conduction states by selectively setting the conduction
state of the material. The conduction state of the memristor
material can be read by applying a voltage across the material and
measuring the current that passes through the target RPU cells.
[0014] Although current cross-bar array systems with RPU cells show
great promise as components of computing architectures, these
devices are subject to some inherent disadvantages which detract
from their overall usefulness in ANN applications. For example,
current RPU cells are only two terminal memory devices which suffer
from issues such as leakage drift. The presence of drift affects
the accuracy of the artificial intelligence algorithm executed in
connection with the ANN application, particularly, over increased
time periods. Current drift also results in limited array size and
large power consumption, which significantly degrades performance
of an array. In addition, non-linearity issues associated with
conventional RPU circuits impact programming of the RPU cells.
[0015] Accordingly, the present invention is directed to a
cross-bar array system having multi-terminal RPU cells configured
to mitigate drift and improve linearity. In illustrative
embodiments, the cross-bar array system incorporates multiple rails
for each level coupled to a single RPU cell whereby the RPU cell,
in effect, is a four-terminal device.
[0016] Referring now to FIG. 1, one illustrative neuromorphic or
computer architecture that incorporates the cross-bar array system
of the present disclosure is illustrated. FIG. 1 is a diagram of a
two-dimensional (2D) cross-bar array that performs forward matrix
multiplication, backward matrix multiplication and weight updates
according to the present disclosure. The exemplative neuromorphic
device architecture 100 comprises data interfaces 102-1 and 102-2,
peripheral circuits 104-1 and 104-2, conductive row control lines
or rails 106-1, 106-2 . . . 106-N, and conductive column control
lines or rails 108-1, 108-2 . . . 108-M. A matrix of RPU cells
1101,1, 110.sub.1,2 . . . 110.sub.1,M, 110.sub.2,1, 110.sub.2,2 . .
. 110.sub.2,M, . . . 110.sub.N,1, 110.sub.N,2 . . . 110.sub.N,M is
connected to the row control rails 106 and column control rails 108
where each RPU cell is connected to one of the row control rails
106 and one of the column control rails 108. For example, RPU cell
110.sub.1,1 is connected to row control rail 106-1 and column
control rail 108-1. Such an example architecture is sometimes
referred to as a 2D cross-bar array. In some embodiments,
neuromorphic device architecture 100 may also comprise reference
circuitry (not shown) which comprises various types of circuit
blocks such as power, clock, bias and timing circuitry to provide
power distribution, control signals and clocking signals for
operation of the neuromorphic device architecture 100.
[0017] In some embodiments, the neuromorphic device architecture
100 is configured to perform ANN computations wherein a conductance
of each RPU cell 110 represents a matrix element or weight, which
can be updated or accessed through operations of the peripheral
circuits 104. Training generally relies on a backpropagation
process which comprises three repeating cycles: a forward cycle, a
backward cycle, and a weight update cycle. The neuromorphic device
architecture 100 can be configured to perform all three cycles of
the backpropagation process in parallel, thus potentially providing
significant acceleration in training with lower power and reduced
computation resources.
[0018] The programmable resistive RPU cells provide local data
storage functionality and local data processing functionality. In
other words, when performing data processing, the value stored at
each RPU is permanently stored, which eliminates the need to move
relevant data in and out of a processor and a separate storage
element. Additionally, the local data storage and local data
processing provided by the described embodiments accelerate the
ANN' s ability to implement algorithms such as matrix inversion,
matrix decomposition and the like. Accordingly, implementing a
machine learning ANN architecture having the described RPU enables
the implementation that optimizes speed, efficiency and power
consumption of the ANN. The described RPU and resulting ANN
architecture improve overall ANN performance and enable a broader
range of practical ANN applications.
[0019] In some embodiments, the RPU cells 110 can be implemented
using variable resistance devices or memristor devices such as
resistive random-access memory (RRAM) devices, phase change memory
(PCM) devices, programmable cell (PMC) memory devices, etc. In
other embodiments, the RPU cells 110 can be implemented using an
analog CMOS framework comprising a capacitor and one or more read
transistors, wherein the capacitor serves as a memory element in
the RPU cell which stores a weight value in the form of capacitor
voltage. The capacitor voltage is directly applied to the gate
terminal(s) of the read transistor(s) to modulate the channel
resistance of the read transistor(s). The charge state stored in
the capacitor can be accessed by applying a small bias across the
read transistor(s) to generate drain currents, which generate a
read current that flows out of RPU cell. The RPU cells are
configured to implement algorithms, for example artificial
intelligence (AI) algorithms, or other functions. In applications,
fast and scalable architectures for matrix operations (e.g.,
inversion, multiplications, etc.) with the RPU cells can be
achieved.
[0020] In FIG. 1, the conductive row control rails 106 and the
conductive column control rails 108 are each shown in FIG. 1 as a
single line for ease of initial illustration; however, in
accordance with exemplative embodiments of the present disclosure,
at least some of the row control rails and the column control rails
include two or more individual lines with a first of the individual
line performing a write/update function and a second of an
individual line performing a read function. More specifically, in
illustrative embodiments, each row control rail 106 and each column
control rail 108 comprises a pair of control lines, including a
"read" control line, and an "update/write" control line, wherein
the cell conductance (e.g., weight value) of a given RPU cell 110
can be adjusted and sensed through the write/update and read
control lines, respectively, for a given row and column connected
to the given RPU cell 110. Furthermore, in conjunction with the
multiline approach for read and write functionalities, the
cross-bar array system of the present disclosure is configured to
provide multiterminal RPU cells, e.g., at least one or more RPU
cells has four terminals. These features provide significant
advantages as will be discussed in further detail hereinbelow.
[0021] Referring now to FIG. 2, one illustrative embodiment of the
cross-bar array system for use with the neuromorphic or computer
architecture 100 represented in FIG. 1 of the present disclosure is
illustrated. The cross-bar array system 200 includes a plurality or
set of conductive row rails 202, a plurality or set of conductive
column rails 204 and a matrix of RPU cells 206 at vertical
cross-points of the row and column rails 202, 204. The conductive
row rails 202 and the conductive column rails 204 are arranged in
general orthogonal relation. Each conductive row rail 202 includes
a first row line 208 and a second row line 210 arranged in parallel
relation and in electrical contact with upper surfaces of
respective RPU cells 206. In illustrative embodiments, the first
row line 208 is a write/modify line (hereinafter, "row write line
208") and the second row line 210 is a read line (hereinafter "row
read line 210"). Similarly, each conductive column rail 204
includes a first column line 212 and a second column line 214
arranged in parallel relation and in electrical contact with the
lower surface of respective RPU cells 206. In illustrative
embodiments, the first column line 212 is a write/modify line
(hereinafter, "column write line 212") and the second row line 214
is a read line (hereinafter "column read line 214"). The cross-bar
system 200 further includes a transistor 216 electrically coupled
to each of the row write and row read lines 208, 210 and to each of
the column write and column read lines 212, 214. In illustrative
embodiments, the transistors 216 may be nMOS transistors, but pMOS
transistors are also envisioned.
[0022] In FIG. 2, only three row rails 202 and three column rails
204 are disclosed. It is appreciated that the cross-bar system 200
may have hundreds of row and column rails 202, 204 with associated
RPU cells 206. One or more of the rows and column rails 202, 204
may include the pairs of respective row write and read lines 208,
210 and the column write and column read lines 212, 214. In
illustrative embodiments, all of the row and column rails 202, 204
include the pairs of write and read lines. In addition, the
cross-bar array system 200 is depicted as a two (2) dimensional
array. It is envisioned that the array may be multidimensional
including three (3) dimensional stacking. The RPU cell 206 of the
cross-bar system 200 provides a four terminal device defined by the
respective couplings of the row write and row read lines 208, 210
and the column write and column read lines 212, 214 with respective
ones of the RPU cells 206. The four terminal device may provide
enhanced flexibility to the computer architecture.
[0023] The use of the cross-bar system 200 of FIG. 2 will now be
described. One or more voltage sources with accompanying logic or
circuitry, represented schematically as "V," are coupled to each of
the row write and read lines 208, 210 and the column write and
column read lines 212, 214, associated with one or more RPU cells
206. In FIG. 2, voltage sources "V" are shown coupled to only one
row rail 202 and one column rail 204 for illustrative purposes. In
applications, one or more voltage sources "V" would be coupled to
each row rail 202 and column rail 204. The voltage characteristic
of the RPU cell 206 is determined. In general, the V.sub.SELECT is
the minimum voltage which if exceeded, i.e., if V>V.sub.SELECT,
would change the internal resistance state of the RPU cell 206. The
V.sub.SELECT may be a positive or negative voltage depending on the
application. In conjunction with performing a write or modify
function to the RPU cell 206, the transistors 216 coupled to each
of the row read line 210 and the column read line 214 are set to an
"off" or "clamp" position (e.g., V=0) preventing current from
passing through these read lines 210, 214. The transistors 216
coupled to the row write line 208 and/or the column write line 212
are set to an "open" state. The voltage sources are set, for
example, to apply a bias application of voltage V>V.sub.SELECT
to the row write line 208 and/or the column write line 212 to
deliver a current through the write lines 208, 212 and perform a
write function on the respective RPU cell 206. No current flows
through the row read and column read lines 210, 214. Similarly, in
conjunction with performing a read function, the transistors 216
associated with the row and/or column write lines 208, 212 are set
to an "off" state to close off the circuit leading to the row and
column write lines 208, 212. The transistors 216 coupled to the row
read line 210 and/or the column read line 214 are set to an "on"
state. A bias voltage, which in illustrative embodiments is 0.1-0.9
V.sub.SELECT, is applied to the row read and column read lines 210,
214 to deliver a current and perform a read function on the
respective RPU cells 206.
[0024] The cross-bar system 200 thus substantially minimizes
cross-talk between the row and column write lines 208, 212 and the
row and column read lines 210, 214. Moreover, the ability to
selectively close the row and column read lines 210, 214 when
performing a write function and close the row and column write
lines 208, 212 when performing a read function, via control of the
respective transistors 216, advantageously reduces any cross-talk
between the lines. This produces an RPU circuit having highly
linear and symmetric programming properties which is significant
for successful implementation in practical ANNs.
[0025] Referring now to FIG. 3, another illustrative embodiment of
the present disclosure is illustrated. In accordance with this
illustrative embodiment, a pMOS transistor 218 is coupled to the
row write line 208 and an nMOS transistor 220 is coupled to the row
read line 210. An nMOS transistor 220 is coupled to the column
write line 212 and a pMOS transistor 218 is coupled to the column
read line 214, i.e., an opposite setup to the row write and row
read lines 208, 210. In addition, a single voltage source "V" is
coupled to the row write and row read lines 208, 210 of the row
rail 202 and a single voltage source "V" is coupled to the column
write and column read lines 212, 214 of the column rail 204. In
this illustrative embodiment, the voltage selection options
include:
[0026] 1. V.sub.SELECT=0--to close off the respective pMOS
transistors 218 or nMOS transistors 220;
[0027] 2. V.sub.SELECT>0--to perform a write function to the RPU
cell 206; and
[0028] 3. V.sub.SELECT<0--to perform a read function on the RPU
cell 206.
[0029] When performing a write function to the RPU cell 206, the
V.sub.SELECT of the voltage source "V" coupled to the row write and
read lines 208, 210 is selected to be a positive V+ voltage source.
Thus, when V.sub.SELECT is positive, the nMOS transistor 220 is in
a closed state to prevent current flow through the row read line
210. In turn, the pMOS transistor 218 assumes an open state
permitting current to flow through the row write line 208 to
perform a write function on the RPU cell 206.
[0030] As noted hereinabove, the column rail 204 has the opposite
transistor setup. The column write line 212 is coupled to an nMOS
transistor 220 while the column read line 214 is coupled to a pMOS
transistor 218. When VSELECT is positive, the nMOS transistor 220
assumes an open state to allow current to flow through the column
write line 212, while the pMOS transistor 218 is in a closed state
to clamp-off the column read line 214. In this manner, with a
positive V.sub.SELECT, the column write line 212 will allow current
flow and drop the voltage across the write electrodes to perform a
write function on the RPU cell 206 while no current will flow
through the column read line 214.
[0031] When conducting a read function on the RPU cell 206, the
V.sub.SELECT of the voltage source "V" to either the row rail 202
and/or the column rail 204 is selected to be a negative V- voltage
source. With a negative VSELECT applied to the row rail 202, the
nMOS transistor 220 coupled to the row read line 210 assumes an
open state to allow current flow to perform a read function on the
RPU cell 206. The pMOS transistor 218 assumes a closed state.
Similarly, with a negative V.sub.SELECT applied to the column rail
204, the pMOS transistor 218 coupled to the column read line 214
assumes an open state to allow current flow to perform a read
function on the RPU cell 206. The nMOS transistor 220 coupled to
the column write line 212 will assume a closed state. It is noted
that all other row and column rails lines should have
V.sub.SELECT=0, which will clamp those lines and ensure no current
flows through unselected lines/devices. In an alternative
embodiment, the pMOS and nMOS transistors 218, 220 respectively
coupled to the row write and row read lines 208, 210 may be
switched, and the nMOS and pMOS transistors 220, 218 respectively
coupled to the column write and column read lines 212, 214 may be
switched. The V.sub.SELECT will also be switched whereby a negative
V.sub.SELECT results in a write function performed on the RPU cell
206 and a positive V.sub.SELECT results in a read function on the
RPU cell 206.
[0032] Referring now to FIG. 4, another illustrative embodiment of
the present disclosure is illustrated. Similar to the prior
illustrative embodiments, the cross-bar system 300 includes row and
column rails 302, 304 interconnected by RPU cells 306. Each row
rail 302 includes a row write line 308 and a row read line 310
while each column rail 304 includes a column write line 312 and a
column read line 314. However, the write lines 308, 312 and the
read lines 310, 314 comprise different materials. More
specifically, the row and column write lines 308, 312 are
fabricated from a material having a greater resistivity than the
material from which the row read line and the column read lines
310, 314 are fabricated. For example, the row and column write
lines 308, 312 may comprise titanium nitride (TiN) and/or tantalum
nitride (TaN) while the row read and column read lines 310, 314 may
comprise copper (Cu), cobalt (Co), ruthenium (Ru), or tungsten (W).
In addition, an nMOS transistor 316 is coupled to each of the row
read line 310 and column read line 314. One or more voltage sources
"V" with designated circuitry and logic is coupled to each of the
row write line 308, row read line 310, column write line 312 and
column read line 314.
[0033] In conjunction with performing a write/modify function to
the RPU cells 306, the nMOS transistor 316 coupled to each of the
row read line 310 and the column read line 314 are in a closed
state to correspondingly close or "clamp down" the row and column
read lines 310, 314, i.e., no voltage is supplied to the nMOS
transistors 316. A voltage, via the voltage sources V, is supplied
to the row and column write lines 308, 312 and a current, at the
appropriate levels, is carried through these lines 308, 312 to
write/modify the respective RPU cells 306.
[0034] In conjunction with performing a read function, a positive
voltage is supplied from each voltage source "V" to the nMOS
transistors 316 to open the nMOS transistors 316. A resulting
current, at specified levels controlled through the logic or
circuitry associated with the voltage sources "V," is conveyed
along the row and column read lines 310, 314 to perform one or more
read functions on the respective RPU cell 306. Due to the relative
high resistivity of the material of fabrication of the row write
and column write lines 308, 312, cross-talk current between the
read lines 310, 314 and the write lines 308, 312 is minimized. More
specifically, the high resistance material of the row and column
write lines 308, 312 substantially impedes current flow or drift
along these lines 308, 312, which would otherwise occur as a
product of cross-talk between the read lines 310, 314 and the write
lines 308, 312.
[0035] It is to be understood that the embodiments discussed herein
are not limited to the particular materials, features, and
processing steps shown and described herein. In particular, with
respect to the processing steps, it is to be emphasized that the
descriptions provided herein are not intended to encompass all of
the processing steps that may be required to form a functional
semiconductor integrated circuit device.
[0036] It is to be understood that the methods discussed herein can
be incorporated in various semiconductor process flows to fabricate
cross-bar array structures (or other memory arrays which comprise
resistive memory cells that require electroforming), in conjunction
with integrated circuits having analog and digital circuitry or
mixed-signal circuitry. In particular, integrated circuit dies can
be fabricated with various devices such as FinFET devices, bipolar
transistors, metal-oxide-semiconductor transistors, diodes,
capacitors, inductors, etc. An integrated circuit can be employed
in applications, hardware, and/or electronic systems. Suitable
hardware and systems for implementing the invention may include,
but are not limited to, personal computers, communication networks,
electronic commerce systems, portable communications devices (e.g.,
cell phones), solid-state media storage devices, functional
circuitry, etc. Systems and hardware incorporating such integrated
circuits are considered part of the embodiments described herein.
Given the teachings of the invention provided herein, one of
ordinary skill in the art will be able to contemplate other
implementations and applications of the techniques described herein
according to embodiments of the invention.
[0037] Although exemplary embodiments have been described herein
with reference to the accompanying figures, it is to be understood
that the invention is not limited to those precise embodiments, and
that various other changes and modifications may be made therein by
one skilled in the art without departing from the scope of the
appended claims.
* * * * *