Core Concepts
A novel methodology for mapping Ising optimization problems to artificially increase the effective precision of low-precision Ising solvers, enabling them to achieve high-precision performance without hardware changes.
Abstract
The paper presents a novel approach to increase the effective precision of Ising solvers, which are hardware implementations of Ising optimization problems. Real-world Ising solvers often have limited precision in representing the problem coefficients, which can significantly impact their performance compared to simulated models.
The key insights are:
Ising solvers can typically represent coefficients in the range [-Cmax, Cmax], limiting the precision.
The authors propose a multi-digit Ising mapping technique to artificially increase the effective precision.
In the 3-digit base-q representation, each coefficient is represented using three base-q digits, effectively increasing the range to [-Mq, Mq], where Mq = (q-1)(q^2 + q + 1).
For Ising solvers with limited spins, a 2-digit base-q representation is proposed, which can still significantly improve the effective precision.
The authors evaluate the proposed techniques on the COBI Ising solver for MIMO signal detection, demonstrating substantial improvements in bit error rate performance compared to the native low-precision mapping.
Stats
The paper does not provide any explicit numerical data or statistics. The key results are presented through performance plots comparing the bit error rate of the proposed multi-digit Ising mapping against the native low-precision mapping for MIMO signal detection scenarios.