Abstract:Recent works demonstrated the promise of using resistive random access memory (ReRAM) as an emerging technology to perform inherently parallel analog domain in-situ matrix-vector multiplication -- the intensive and key computation in DNNs. With weights stored in the ReRAM crossbar cells as conductance, when the input vector is applied to word lines, the matrix-vector multiplication results can be generated as the current in bit lines. A key problem is that the weight can be either positive or negative, but the in-situ computation assumes all cells on each crossbar column with the same sign. The current architectures either use two ReRAM crossbars for positive and negative weights, or add an offset to weights so that all values become positive. Neither solution is ideal: they either double the cost of crossbars, or incur extra offset circuity. To better solve this problem, this paper proposes FORMS, a fine-grained ReRAM-based DNN accelerator with polarized weights. Instead of trying to represent the positive/negative weights, our key design principle is to enforce exactly what is assumed in the in-situ computation -- ensuring that all weights in the same column of a crossbar have the same sign. It naturally avoids the cost of an additional crossbar. Such weights can be nicely generated using alternating direction method of multipliers (ADMM) regularized optimization, which can exactly enforce certain patterns in DNN weights. To achieve high accuracy, we propose to use fine-grained sub-array columns, which provide a unique opportunity for input zero-skipping, significantly avoiding unnecessary computations. It also makes the hardware much easier to implement. Putting all together, with the same optimized models, FORMS achieves significant throughput improvement and speed up in frame per second over ISAAC with similar area cost.

Polar Code Construction for Resistive Memories With Sneak-Path Interference

Performance Limit and Coding Schemes for Resistive Random-Access Memory Channels

Non-Stationary Polar Codes for Resistive Memories

Across-Array Coding for Resistive Memories with Sneak-Path Interference and Lognormal Distributed Resistance Variations

Sneak Path Interference-Aware Adaptive Detection and Decoding for Resistive Memory Arrays

Performance Analysis and Code Design for Resistive Random-Access Memory Using Channel Decomposition Approach

Constrained Coding and Deep Learning Aided Threshold Detection for Resistive Memories

Circuit Modeling for RRAM-Based Neuromorphic Chip Crossbar Array with and Without Write-Verify Scheme.

Data-Driven Neural Polar Codes for Unknown Channels With and Without Memory

Polar Codes with Memory

Quantization Design for Resistive Memories With Multiple Reads

Cross-layer Designs against Non-ideal Effects in ReRAM-based Processing-in-Memory System

Maximum Achievable Rate of Resistive Random-Access Memory Channels by Mutual Information Spectrum Analysis

Efficient LDPC Code Design for Combating Asymmetric Errors in STT-RAM.

An Efficient Construction Method Based on Partial Distance of Polar Codes with Reed-Solomon Kernel

Protograph QC-LDPC and Rate-Adaptive Polar Codes Design for MLC NAND Flash Memories.

Polar-Coded Forward Error Correction for MLC NAND Flash Memory Polar FEC for NAND Flash Memory

Efficient Low-Density Parity-Check (LDPC) Code Decoding for Combating Asymmetric Errors in STT-RAM

Self-Terminating Write of Multi-Level Cell ReRAM for Efficient Neuromorphic Computing

FORMS: Fine-grained Polarized ReRAM-based In-situ Computation for Mixed-signal DNN Accelerator