Lookup Table-based Multiplication-free All-digital DNN Accelerator Featuring Self-Synchronous Pipeline Accumulation

2025/06/20 by Tagata, Hiroto, Sato, Takashi, Awano, Hiromitsu
Computer Science · Engineering · #FOS: Computer and information sciences #Hardware Architecture (cs.AR) #Low-power high-performance VLSI design #Numerical Methods and Algorithms #Parallel Computing and Optimization Techniques

paper · pdf · doi:10.48550/arxiv.2506.16800

openalex publication_date 2025/06/20 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Deep neural networks (DNNs) have been widely applied in our society, yet reducing power consumption due to large-scale matrix computations remains a critical challenge. MADDNESS is a known approach to improving energy efficiency by substituting matrix multiplication with table lookup operations. Previous research has employed large analog computing circuits to convert inputs into LUT addresses, which presents challenges to area efficiency and computational accuracy. This paper proposes a novel MADDNESS-based all-digital accelerator featuring a self-synchronous pipeline accumulator, resulting in a compact, energy-efficient, and PVT-invariant computation. Post-layout simulation using a commercial 22nm process showed that 2.5 times higher energy efficiency (174 TOPS/W) and 5 times higher area efficiency (2.01 TOPS/mm2) can be achieved compared to the conventional accelerator.

Citations

• OLLA: Optimizing the Lifetime and Location of Arrays to Reduce the Memory Usage of Neural Networks
• PECAN: A Product-Quantized Content Addressable Memory Network
• Look-ups are not (yet) all you need for deep learning inference
• Multiplying Matrices Without Multiplying
• BiQGEMM: Matrix Multiplication with Lookup Table For Binary-Coding-based Quantized DNNs
• Deep Residual Learning for Image Recognition
• Training Very Deep Networks
• ImageNet classification with deep convolutional neural networks
• Stella Nera: A Differentiable Maddness-Based Hardware Accelerator for Efficient Approximate Matrix Multiplication
• Gradient-based learning applied to document recognition

Related

• HADES: Hardware/Algorithm Co-design in DNN accelerators using Energy-efficient Approximate Alphabet Set Multipliers
• Energy-efficient Dense DNN Acceleration with Signed Bit-slice Architecture
• Heterogeneous Multi-core Array-based DNN Accelerator
• An Energy-Efficient Near-Data Processing Accelerator for DNNs that Optimizes Data Accesses
• NL-DPE: An Analog In-memory Non-Linear Dot Product Engine for Efficient CNN and LLM Inference