Research Paper#Hardware Architecture, Combinatorial Optimization, Edge Computing🔬 ResearchAnalyzed: Jan 3, 2026 16:11
LIMO: Low-Power In-Memory Annealer for Edge Computing
Published:Dec 29, 2025 05:20
•1 min read
•ArXiv
Analysis
This paper introduces LIMO, a novel hardware architecture designed for efficient combinatorial optimization and matrix multiplication, particularly relevant for edge computing. It addresses the limitations of traditional von Neumann architectures by employing in-memory computation and a divide-and-conquer approach. The use of STT-MTJs for stochastic annealing and the ability to handle large-scale instances are key contributions. The paper's significance lies in its potential to improve solution quality, reduce time-to-solution, and enable energy-efficient processing for applications like the Traveling Salesman Problem and neural network inference on edge devices.
Key Takeaways
- •LIMO is a mixed-signal computational macro for in-memory annealing.
- •It utilizes STT-MTJs for stochastic annealing to escape local minima.
- •A divide-and-conquer algorithm is used for large instances.
- •LIMO achieves superior solution quality and faster time-to-solution compared to prior hardware annealers.
- •The macro can be reused for vector-matrix multiplications (VMMs) and neural network inference.
Reference
“LIMO achieves superior solution quality and faster time-to-solution on instances up to 85,900 cities compared to prior hardware annealers.”