Qualcomm Incorporated

Yonsei University

Industry-academic Cooperation Foundation

Zhongze Wang

Choh Fei Yeap

Stanley Seungchul Song

Xia Li

Niladri Mojumder

Xiaonan Chen

John Jianhong Zhu

Kern Rim

Seong-Ook Jung

Ping Liu

Jeffrey Xu

Mustafa Badaroglu

Xiaochun Zhu

Ye Lu

Bin Yang

Xiao Lu

Yandong Gao

Younghwi Yang

Edward Harrison Teague

Periannan Chidambaram

Max Welling

Beom-Mo Han

Haining S Yang

Xiangdong Chen

Mohamed Hassan Abu-Rahma

Yu Lu

Ritu Chaba

Seung-Chul Song

Lixin Ge

Hyunkook Park

Guoqing Chen

Seung Hyuk Kang

Matthew Michael Nowak

Changho Jung

Jun Yuan

Sameer Wadhwa

Nan Chen

Ankit Srivastava

Jeffrey Herbert Fischer

Manish Garg

Jonathan H Liu

Ohsang Kwon

Da Yang

Foua Vang

Seyed Arash Mirhaj

Chock Hing Gan

Raymond George Stephany

Paul Simon Hoayun

Daniel Wayne Perry

Matthias Reisser

Sian-Yee Sean Lee

Yong Park

Weidan Li

Saurabh Kedar Pitre

Mingu Kang

Jeffrey Junhao Xu

Niladri Narayan Mojumder

Yang Da

Methods and apparatus for performing machine learning tasks, and in particular, to a neural-network-processing architecture and circuits for improved handling of partial accumulation results in weight-stationary operations, such as operations occurring in compute-in-memory (CIM) processing elements (PEs). One example PE circuit for machine learning generally includes an accumulator circuit, a flip-flop array having an input coupled to an output of the accumulator circuit, a write register, and a first multiplexer having a first input coupled to an output of the write register, having a second input coupled to an output of the flip-flop array, and having an output coupled to a first input of the first accumulator circuit.

Partial sum management and reconfigurable systolic flow architectures for in-memory computation

An integrated circuit includes a trench power rail to reduce resistance in a power rail or avoid an increase in resistance of a power rail as a result of the metal tracks being reduced in size as the technology node size is reduced. The trench power rail is formed in isolation regions between cell circuits. A cell isolation trench in the isolation region provides additional volume in which to dispose additional metal material for forming the trench power rail to increase its cross-sectional area. The trench power rail extends through a via layer to a metal layer, including signal interconnects. The trench power rail extends in a width direction out of the cell isolation trench in the via layer to couple to trench contacts of the adjacent cell circuits without vertical interconnect accesses (vias). A high-K dielectric layer can selectively isolate the trench power rail from the cell circuits.

Trench power rail in cell circuits to reduce resistance and related power distribution networks and fabrication methods

Certain aspects generally relate to performing machine learning tasks, and in particular, to computation-in-memory architectures and operations. One aspect provides a circuit for in-memory computation. The circuit generally includes multiple bit-lines, multiple word-lines, an array of compute-in-memory cells, and a plurality of accumulators, each accumulator being coupled to a respective one of the multiple bit-lines. Each compute-in-memory cell is coupled to one of the bit-lines and to one of the word-lines and is configured to store a weight bit of a neural network.

Digital compute in memory

In one embodiment, a method of simulating an operation of an artificial neural network on a binary neural network processor includes receiving a binary input vector for a layer including a probabilistic binary weight matrix and performing vector-matrix multiplication of the input vector with the probabilistic binary weight matrix, wherein the multiplication results are modified by simulated binary-neural-processing hardware noise, to generate a binary output vector, where the simulation is performed in the forward pass of a training algorithm for a neural network model for the binary-neural-processing hardware.

Systems and methods for modifying neural networks for binary processing applications

A compute-in-memory bitcell is provided that includes a pair of cross-coupled inverters for storing a stored bit. The compute-in-memory bitcell includes a logic gate for multiplying the stored bit with an input vector bit. An output node for the logic gate connects to a second plate of a capacitor. A first plate of the capacitor connects to a read bit line. A write driver controls a power supply voltage to the cross-coupled inverters, the first switch, and the second switch to capacitively write the stored bit to the pair of cross-coupled inverters.

Compute-in-memory bitcell with capacitively-coupled write operation

Compute-in-memory (CIM) bit cell array circuits include CIM bit cell circuits for multiply-accumulate operations. The CIM bit cell circuits include a memory bit cell circuit for storing a weight data in true and complement form. The CIM bit cell circuits include a true pass-gate circuit and a complement pass-gate circuit for generating a binary product of the weight data and an activation input on a product node. An RWL circuit couples the product node to a ground voltage for initialization. The CIM bit cell circuits also include a plurality of consecutive gates each coupled to at least one of the memory bit cell circuit, the true pass-gate circuit, the complement pass-gate circuit, and the RWL circuit. Each of the CIM bit cell circuits in the CIM bit cell array circuit is disposed in an orientation of a CIM bit cell circuit layout including the RWL circuit.

Compute-in-memory (CIM) bit cell circuits each disposed in an orientation of a cim bit cell circuit layout including a read word line (RWL) circuit in a cim bit cell array circuit

Digital compute-in-memory (DCIM) bit cell circuit layouts and DCIM array circuits for multiple operations per column are disclosed. A DCIM bit cell array circuit including DCIM bit cell circuits comprising exemplary DCIM bit cell circuit layouts disposed in columns is configured to evaluate the results of multiple multiply operations per clock cycle. The DCIM bit cell circuits in the DCIM bit cell circuit layouts each couples to one of a plurality of column output lines in a column. In this regard, in each cycle of a system clock, each of the plurality of column output lines receives a result of a multiply operation of a DCIM bit cell circuit coupled to the column output line. The DCIM bit cell array circuit includes digital sense amplifiers coupled to each of the plurality of column output lines to reliably evaluate a result of a plurality of multiply operations per cycle.

Digital compute-in-memory (DCIM) bit cell circuit layouts and DCIM arrays for multiple operations per column

An apparatus includes first and second compute-in-memory (CIM) arrays. The first CIM array is configured to store weights corresponding to a filter tensor, to receive a first set of activations corresponding to a first receptive field of an input, and to process the first set of activations with the weights to generate a corresponding first tensor of output values. The second CIM array is configured to store a first copy of the weights corresponding to the filter tensor and to receive a second set of activations corresponding to a second receptive field of the input. The second CIM array is also configured to process the second set of activations with the first copy of the weights to generate a corresponding second tensor of output values. The first and second compute-in-memory arrays are configured to process the first and second receptive fields in parallel.

Parallel processing of a convolutional layer of a neural network with compute-in-memory array

A method performs XNOR-equivalent operations by adjusting column thresholds of a compute-in-memory array of an artificial neural network. The method includes adjusting an activation threshold generated for each column of the compute-in-memory array based on a function of a weight value and an activation value. The method also includes calculating a conversion bias current reference based on an input value from an input vector to the compute-in-memory array, the compute-in-memory array being programmed with a set of weights. The adjusted activation threshold and the conversion bias current reference are used as a threshold for determining the output values of the compute-in-memory array.

Performing XNOR equivalent operations by adjusting column thresholds of a compute-in-memory array

In one embodiment, an electronic device includes a compute-in-memory (CIM) array that includes a plurality of columns. Each column includes a plurality of CIM cells connected to a corresponding read bitline, a plurality of offset cells configured to provide a programmable offset value for the column, and an analog-to-digital converter (ADC) having the corresponding bitline as a first input and configured to receive the programmable offset value. Each CIM cell is configured to store a corresponding weight.

Voltage offset for compute-in-memory architecture

Certain aspects provide a circuit for in-memory computation. The circuit generally includes an in-memory computation array having a plurality of computation circuits, each of the computation circuits being configured to perform a dot product computation. In certain aspects, each of the computation circuits includes a memory cell, a capacitive element, a precharge transistor coupled between an output of the memory cell and the capacitive element, and a read transistor coupled between a read bit line (RBL) and the capacitive element.

Memory cell for dot product operation in compute-in-memory chip

Certain aspects provide a circuit for in-memory computation. The circuit generally includes a first memory cell, and a first computation circuit. The first computation circuit may include a first switch having a control input coupled to an output of the first memory cell, a second switch coupled between a node of the first computation circuit and the first switch, a control input of the second switch being coupled to a discharge word-line (DCWL), a capacitive element coupled between the node and a reference potential node, a third switch coupled between the node and a read bit-line (RBL), and a fourth switch coupled between the node and an activation (ACT) line.

Charge-sharing compute-in-memory system

A bit cell circuit of a most-significant bit (MSB) of a multi-bit product generated in an array of bit cells in a compute-in-memory (CIM) array circuit is configured to receive a higher supply voltage than a supply voltage provided to a bit cell circuit of another bit cell corresponding to another bit of the multi-bit product. A bit cell circuit receiving a higher supply voltage increases a voltage difference between increments of an accumulated voltage, which can increase accuracy of an analog-to-digital converter determining a pop-count. A bit cell circuit of the MSB in the CIM array circuit receives the higher supply voltage to increase accuracy of the MSB which increases accuracy of the CIM array circuit output. A capacitance of a capacitor in the bit cell circuit of the MSB is smaller to avoid an increase in energy consumption due to the higher voltage.

Multi-bit compute-in-memory (CIM) arrays employing bit cell circuits optimized for accuracy and power efficiency

Certain aspects provide methods and apparatus for multiplication of digital signals. In accordance with certain aspects, a multiplication circuit may be used to multiply a portion of a first digital input signal with a portion of a second digital input signal via a first multiplier circuit to generate a first multiplication signal, and multiply another portion of the first digital input signal with another portion of the second digital input signal via a second multiplier circuit to generate a second multiplication signal. A third multiplier circuit and multiple adder circuits may be used to generate an output of the multiplication circuit based on the first and second multiplication signals.

Fast digital multiply-accumulate (MAC) by fast digital multiplication circuit

Certain aspects provide methods and apparatus for binary computation. An example circuit for such computation generally includes a memory cell having at least one of a bit-line or a complementary bit-line; a computation circuit coupled to a computation input node of the circuit and the bit-line or the complementary bit-line; and an adder coupled to the computation circuit, wherein the computation circuit comprises a first n-type metal-oxide-semiconductor (NMOS) transistor coupled to the memory cell, and a first p-type metal-oxide-semiconductor (PMOS) transistor coupled to the memory cell, drains of the first NMOS and PMOS transistors being coupled to the adder, wherein a source of the first PMOS transistor is coupled to a reference potential node, and wherein a source of the first NMOS transistor is coupled to the computation input node.

Compute-in-memory (CIM) binary multiplier

Growing community of inventors

San Diego, CA, United States of America

Zhongze Wang