Recent Advances in Compute-in-Memory Support for SRAM Using Monolithic 3-D Integration

Zhang, Zhixiao; Si, Xin; Srinivasa, Srivatsa; Ramanathan, Akshay Krishna; Chang, Meng‐Fan

doi:10.1109/mm.2019.2946489

Cited by 6 publications

(3 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Moreover, the forward voltage of diode D2 𝑉 minus the forward voltage of diode D1 𝑉 is equal to 𝑉 , which is because of the virtual short characteristic of the operational amplifier. Consequently, Equation ( 6) can be further simplified as (7), 𝐼 𝐼…”

Section: Peripheral Circuitsmentioning

confidence: 99%

“…A variety of works aim to solve this problem. Among them, the Computing-In-Memory (CIM) [4][5][6][7] architecture was spotlighted because of its extraordinary advantages, one is the great reduction of the data transmission, while the other is the substantially improved parallelism. Plenty of previous works have demonstrated the design methods and results of CIM, which greatly showed the superiority of CIM architecture [8][9][10][11][12][13].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A High-Precision Implementation of the Sigmoid Activation Function for Computing-in-Memory Architecture

Xie

et al. 2021

Micromachines

View full text Add to dashboard Cite

Computing-In-Memory (CIM), based on non-von Neumann architecture, has lately received significant attention due to its lower overhead in delay and higher energy efficiency in convolutional and fully-connected neural network computing. Growing works have given the priority to researching the array of memory and peripheral circuits to achieve multiply-and-accumulate (MAC) operation, but not enough attention has been paid to the high-precision hardware implementation of non-linear layers up to now, which still causes time overhead and power consumption. Sigmoid is a widely used non-linear activation function and most of its studies provided an approximation of the function expression rather than totally matched, inevitably leading to considerable error. To address this issue, we propose a high-precision circuit implementation of the sigmoid, matching the expression exactly for the first time. The simulation results with the SMIC 40 nm process suggest that the proposed circuit implemented high-precision sigmoid perfectly achieves the properties of the ideal sigmoid, showing the maximum error and average error between the proposed simulated sigmoid and ideal sigmoid is 2.74% and 0.21%, respectively. In addition, a multi-layer convolutional neural network based on CIM architecture employing the simulated high-precision sigmoid activation function verifies the similar recognition accuracy on the test database of handwritten digits compared to utilize the ideal sigmoid in software, with online training achieving 97.06% and with offline training achieving 97.74%.

show abstract

Section: Peripheral Circuitsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

A High-Precision Implementation of the Sigmoid Activation Function for Computing-in-Memory Architecture

Xie

et al. 2021

Micromachines

View full text Add to dashboard Cite

show abstract

“…[1][2][3] It integrates computing and storage cell in parallel, effectively eliminating the need for frequent data transformation between memory and processing units in the von Neumann architecture, holding immense potential for a wide range of applications. [4][5][6][7][8] In CIM, parallel data processing can be achieved through vector-matrix multiplication using Ohm's law (for multiplication) and Kirchhoff 's law (for accumulation) in a crossbar array structure. [9][10][11] Synaptic devices are employed to accurately copy the synaptic plasticity in the biological brain, underlying the precise vector-matrix multiplication.…”

Section: Introductionmentioning

confidence: 99%

Electrolyte‐Gated Transistor Array (20 × 20) with Low‐Programming Interference Based on Coplanar Gate Structure for Unsupervised Learning

Zhang,

Li,

et al. 2024

Small Science

View full text Add to dashboard Cite

Compute‐in‐memory (CIM) is a pioneering approach using parallel data processing to eliminate traditional data transmission bottlenecks for faster, energy‐efficient data handling. Crossbar arrays with two‐terminal devices such as memristors and phase‐change memory are commonly employed in CIM, but they encounter challenges such as leakage current and increased power usage. Three‐terminal transistor arrays have potential solutions, yet large‐scale electrolyte‐gated transistors (EGTs) demonstrations are uncommon due to compatibility issues with existing photolithography processes. Herein, a 20 × 20 EGTs array is designed using indium‐gallium‐zinc‐oxide as the semiconductor channel and polyacrylonitrile (PAN) doped with C2F6LiNO4S2 as the electrolyte. Each transistor unit in the array can serve as a synapse, exhibiting a large conductance range, low energy consumption (6.984 fJ) for read–write operations, excellent repeatability, and quasilinear update characteristics. It has been confirmed that the EGTs array not only enables precise device programming but also virtually eliminates signal interference between neighboring devices during the programming process. Using 54 transistors in the EGTs array, unsupervised learning with a winner‐takes‐all neural network is successfully demonstrated. After 50 training iterations, the neural network achieves perfect 100% accuracy in classifying test‐set letters. The work demonstrates the potential of EGTs for constructing large‐scale integration synaptic array toward efficient computing architectures.

show abstract

Emerging monolithic 3D integration: Opportunities and challenges from the computer system perspective

2022

View full text Add to dashboard Cite

Recent Advances in Compute-in-Memory Support for SRAM Using Monolithic 3-D Integration

Cited by 6 publications

References 22 publications

A High-Precision Implementation of the Sigmoid Activation Function for Computing-in-Memory Architecture

A High-Precision Implementation of the Sigmoid Activation Function for Computing-in-Memory Architecture

Electrolyte‐Gated Transistor Array (20 × 20) with Low‐Programming Interference Based on Coplanar Gate Structure for Unsupervised Learning

Emerging monolithic 3D integration: Opportunities and challenges from the computer system perspective

Contact Info

Product

Resources

About