AXELRAM: Quantize Once, Never Dequantize

TL;DR

AXELRAM introduces a fixed codebook architecture for quantized attention computation that significantly reduces multiplications and addresses stability issues with a novel sign pattern selection method.

Contribution

The paper presents a novel SRAM macro architecture enabling direct attention score computation from quantized cache indices without dequantization, improving efficiency and stability.

Findings

Reduces per-query multiplications by 102.4x

Identifies sign pattern sensitivity causing PPL spikes in some models

Proposes a gradient-free sign pattern selection method that eliminates catastrophic spikes

Abstract

We propose AXELRAM, a smart SRAM macro architecture that computes attention scores directly from quantized KV cache indices without dequantization. The key enabler is a design-time fixed codebook: orthogonal-transform-based quantization concentrates each coordinate's distribution to N(0,1/d), so the optimal quantizer depends only on dimension d and bit-width b, not on input data. The asymmetric path design -- transform on write, table-lookup on read with no inverse transform -- reduces per-query multiplications by 102.4x (a mathematical identity). Through multi-seed evaluation (10 seeds x 3 models), we discover that sign pattern sensitivity causes catastrophic PPL spikes (Delta > 50) on certain models (Qwen2.5-3B), while others (LLaMA-3.1-8B) are fully stable. This phenomenon extends SpinQuant's observation of rotation variance in weight quantization to the KV cache domain, where the effect is qualitatively more severe. We trace the root cause to layer-wise norm heterogeneity and propose a gradient-free sign pattern selection (200 candidates, 8 calibration samples, one-time) that eliminates catastrophic spikes with zero additional hardware cost. All source code is available at https://github.com/Axelidea/AXELRAM.