The technology combines memory devices, in-memory computing, and simulated annealing algorithms to create the world＇s first in-memory annealing unit (IMAU). First verified on 10-city traveling salesman problem and 300-read genome assembly, it is expected to be 1000 times faster than current hardware. It serves as a potential tool for high-speed optimal decision-making in chip design, biomedicine, and logistics, presented at IEDM 2021 and awarded first place in the 2024 Micron MIMORY Award.

An in-memory annealing unit (IMAU) has been proposed as an energy-efficient optimizer for the traveling salesman problem and genome assembly. A hardware-algorithm co-optimization approach addresses challenges such as large problem size, weight precision, and analog accuracy. The 1.1Mb RRAM-based IMAU with simulated annealing achieves 90 TOPS and 2000 TOPS/W. We solved 10-city TSP and 300-reads SARS-CoV-2 genome assembly with divide-and-conquer scheme, showing the potential for real applications.

The combinatorial optimization problem (COP) is crucial in biomedicine, transportation, finance, and VLSI routing, requiring substantial computational resources. While quantum annealers are suggested as a solution, they still encounter challenges in cost, complexity, and power consumption. Leveraging commercial semiconductor technology is crucial. The proposed IMAU offers speed, energy efficiency, scalability, and cost-effectiveness with parallelism, analog computing, and reduced data movement.