A Potts Machine With Coefficient Reuse Strategy and Successive Boundary Approximation Annealing for Multi-State Combinatorial Optimization

The annealing processor based on the Ising model has attracted significant attention in recent years due to its remarkable ability in solving combinatorial optimization problems (COPs). By mapping a COP to the annealing processor, the optimal or near-optimal solution can be achieved efficiently. How...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:IEEE journal of solid-state circuits S. 1 - 11
Hauptverfasser: Zhou, Yifeng, Hao, Xin, Cai, Qinchao, Liao, Lei, Chen, Zhuojun
Format: Journal Article
Sprache:Englisch
Veröffentlicht: IEEE 2025
Schlagworte:
Annealing
Annealing processor
CMOS technology
combinatorial optimization problems (COPs)
Computational modeling
Energy efficiency
Integrated circuit modeling
Ising model
Mathematical models
Memory management
Optimization
Potts processor
Semiconductor device modeling
Simulated annealing
ISSN:0018-9200, 1558-173X
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The annealing processor based on the Ising model has attracted significant attention in recent years due to its remarkable ability in solving combinatorial optimization problems (COPs). By mapping a COP to the annealing processor, the optimal or near-optimal solution can be achieved efficiently. However, the Ising model is limited to two spin states, which restricts its application in solving multi-state COPs. Toaddress this limitation, we propose a Potts machine implemented in 65-nm CMOS technology, which is based on a <inline-formula> <tex-math notation="LaTeX">16\times 16</tex-math> </inline-formula> King's graph topology. The proposed Potts machine supports two to four spin states and is compatible with both the Ising model and the Potts model. Additionally, we introduce an intra- and inter-coefficient reuse strategy, which effectively decreases the coefficient memory area overhead by 68.75%. Furthermore, we propose a successive boundary approximation (SBA) annealing method to address the challenges associated with long convergence time. The measurement results demonstrate that the proposed SBA annealing method achieves a <inline-formula> <tex-math notation="LaTeX">10\times </tex-math> </inline-formula> acceleration in comparison to the conventional simulated annealing approach. The proposed Potts machine has a spin area of <inline-formula> <tex-math notation="LaTeX">4400~\mu </tex-math> </inline-formula>m 2 and exhibits an energy consumption of 1.06 nJ when powered by a 1-V supply.
ISSN:0018-9200
1558-173X
DOI:10.1109/JSSC.2025.3630270