Uncertain behaviours of integrated circuits improve computational performance.

Area of Science:

• Computer Science
• Materials Science
• Quantum Computing

Background:

• Semiconductor scaling, the traditional driver of computer performance, is nearing its physical limits.
• New computing paradigms leverage natural phenomena like quantum superposition and stochastic resonance to overcome these limitations.
• Device uncertainties, typically minimized in conventional electronics, are explored as a potential performance enhancer.

Purpose of the Study:

• To investigate whether inherent device uncertainties, arising from semiconductor scaling limitations, can be harnessed to improve computational performance.
• To demonstrate a novel approach to computing that utilizes probabilistic bit errors for enhanced accuracy.

Main Methods:

• Prototyped an integrated circuit designed for ground-state searches of the Ising model.
• Introduced probabilistic bit errors into memory cell devices by manipulating dynamic device characteristics.
• Compared computational performance and solution accuracy with and without these introduced fluctuations.

Main Results:

• Observed significant improvements in solution accuracy for the ground-state search when probabilistic fluctuations were present.
• Demonstrated that intentionally introduced bit errors led to better performance than in devices without such uncertainties.
• Validated the hypothesis that device uncertainties can be a key factor in enhancing computational performance.

Conclusions:

• Uncertain behaviors in semiconductor devices, once considered detrimental, can be strategically utilized to boost computational performance.
• This research opens new avenues for designing future computing architectures that embrace, rather than eliminate, device stochasticity.
• The findings suggest a paradigm shift in electronic device design, moving towards leveraging inherent uncertainties for advanced computation.