Abstract:Deep learning hardware designs have been bottlenecked by conventional memories such as SRAM due to density, leakage and parallel computing challenges. Resistive devices can address the density and volatility issues, but have been limited by peripheral circuit integration. In this work, we demonstrate a scalable RRAM based in-memory computing design, termed XNOR-RRAM, which is fabricated in a 90nm CMOS technology with monolithic integration of RRAM devices between metal 1 and 2. We integrated a 128x64 RRAM array with CMOS peripheral circuits including row/column decoders and flash analog-to-digital converters (ADCs), which collectively become a core component for scalable RRAM-based in-memory computing towards large deep neural networks (DNNs). To maximize the parallelism of in-memory computing, we assert all 128 wordlines of the RRAM array simultaneously, perform analog computing along the bitlines, and digitize the bitline voltages using ADCs. The resistance distribution of low resistance states is tightened by write-verify scheme, and the ADC offset is calibrated. Prototype chip measurements show that the proposed design achieves high binary DNN accuracy of 98.5% for MNIST and 83.5% for CIFAR-10 datasets, respectively, with energy efficiency of 24 TOPS/W and 158 GOPS throughput. This represents 5.6X, 3.2X, 14.1X improvements in throughput, energy-delay product (EDP), and energy-delay-squared product (ED2P), respectively, compared to the state-of-the-art literature. The proposed XNOR-RRAM can enable intelligent functionalities for area-/energy-constrained edge computing devices.

The Impact of Non-linear NVM Devices on In-Memory Computing

Nonlinear Quantization for In-Sram Multi-Bit MAC Design

In-memory Computing with Emerging Nonvolatile Memory Devices

Intra-array Non-Idealities Modeling and Algorithm Optimization for RRAM-based Computing-in-Memory Applications

Compensation Architecture to Alleviate Noise Effects in RRAM-based Computing-in-memory Chips with Residual Resource

High-Throughput In-Memory Computing for Binary Deep Neural Networks with Monolithically Integrated RRAM and 90nm CMOS

Highly Efficient Neuromorphic Computing Systems with Emerging Nonvolatile Memories

Resistive Memory‐Based In‐Memory Computing: From Device and Large‐Scale Integration System Perspectives

Non-volatile Memory Based In-Memory Computing Technology

Non-Volatile Memory for Data Storage and Neuromorphic Computing

Emerging Nonvolatile Memories for Machine Learning

An Overview of In-memory Processing with Emerging Non-volatile Memory for Data-intensive Applications

RRAM-based Coprocessors for Deep Learning

Compensation Architecture Design Utilizing Residual Resource to Mitigate Impacts of Nonidealities in RRAM-based Computing-in-memory Chips

Special Topic on Nonvolatile Memory for Efficient Implementation of Neural/Neuromorphic Computing

In-Memory Logic Operations and Neuromorphic Computing in Non-Volatile Random Access Memory

Accurate Program/Verify Schemes of Resistive Switching Memory (RRAM) for In-Memory Neural Network Circuits

A Co-design view of Compute in-Memory with Non-Volatile Elements for Neural Networks

Emerging Non-Volatile Memories for Computation-in-Memory

Flash Memory Array for Efficient Implementation of Deep Neural Networks

Memory That Never Forgets: Emerging Nonvolatile Memory and the Implication for Architecture Design