This is not the latest version of this item. The latest version can be found here.
Stella Nera: A Differentiable Maddness-Based Hardware Accelerator for Efficient Approximate Matrix Multiplication
METADATA ONLY
Loading...
Author / Producer
Date
2025
Publication Type
Conference Paper
ETH Bibliography
yes
METADATA ONLY
Data
Rights / License
Abstract
Artificial intelligence has surged in recent years, with advancements in machine learning rapidly impacting nearly every area of life. However, the growing complexity of these models has far outpaced advancements in available hardware accelerators, leading to significant computational and energy demands, primarily due to matrix multiplications, which dominate the compute workload. MADDNESS (i.e., Multiply-ADDitioN-lESS) presents a hash-based version of product quantization, which renders matrix multiplications into lookups and additions, eliminating the need for multipliers entirely. We present STELLA NERA1, the first MADDNESS-based accelerator achieving an energy efficiency of 161 TOp/s/W@0.55V, 25x better than conventional MatMul accelerators due to its small components and reduced computational complexity. We further enhance MADDNESS with a differentiable approximation, allowing for gradient-based fine-tuning and achieving an end-to-end performance of 92.5% Top-1 accuracy on CIFAR-10
Permanent link
Publication status
Editor
Book title
2025 IEEE Computer Society Annual Symposium on VLSI (ISVLSI)
Journal / series
Volume
Pages / Article No.
11130225
Publisher
IEEE
Event
28th IEEE Computer Society Annual Symposium on VLSI (ISVLSI 2025)
Edition / version
Methods
Software
Geographic location
Date collected
Date created
Subject
Hardware acceleration; Approximate MatMul; AI
Organisational unit
03996 - Benini, Luca / Benini, Luca