STATICA: A 512-Spin 0.25M-Weight Full-Digital Annealing Processor with a Near-Memory All-SpinUpdates-at-Once Architecture for Combinatorial Optimization with Complete Spin-Spin Interactions

📋 论文概要

该论文提出了一款名为STATICA的全数字退火处理器，集成了512个自旋和0.25M权重，采用近内存全自旋同时更新架构，旨在解决组合优化问题中的计算瓶颈。通过近似分段线性函数替代查找表，实现了轻量级自旋更新单元，提升了能效和速度。

💡 主要创新点

• 提出近内存全自旋同时更新架构，减少数据搬运开销，提高并行处理效率。
• 采用分段线性函数近似sigmoid函数，替代传统查找表，降低自旋更新单元硬件复杂度。
• 实现512自旋和0.25M权重的全数字集成，支持大规模组合优化问题的高效求解。

🏷 关键词

📄 原文摘要

Masanao Yamaoka3, Hiroshi Teramoto2, Akira Sakai2, Shinya Takamaeda-Yamazaki4, Masato Motomura1 Tokyo Institute of Technology, Yokohama, Japan Hokkaido University, Sapporo, Japan, 3Hitachi, Sapporo, Japan 4 University of Tokyo, Tokyo, Japan 1 2 block, respectively. Then they are simultaneously summed, doubled, and accumulated on top of the 𝐿ys from the prior MC step (Fig. 7.3.2 right). To realize light-weight SUUs, STATICA approximates the sigmoidal Py(𝐿y) by a piecewise linear function (Fig. 7.3.4). For 𝜎y ⟶ 𝜏y, while a conventional approach would require a look-up table (LUT) whose output is compared with a uniform [0, 1] random number, our approach allows for discarding the LUT (128KB) and directly comparing 𝐿y𝜎y + 𝑞𝑇 with a ["2𝑇, 2𝑇 ] random number. We share one 32b RNG among two neighboring SUUs and apply 2𝑇-masking for scaling the two 16b independent random numbers to ["2𝑇, 2𝑇 ]. Altogether, this reduces the SUU

👥 作者与机构

Kasho Yamamoto1,2, Kota Ando1, Normann Mertig3, Takashi Takemoto3,