Imprecise Computing

The rise of artificial intelligence, machine learning, and the exponential growth in quantity and scale of data and sensor inputs redefines storage and processing criticality. Optimal task prioritization is morphing from linear-stacked to dynamic-flat. For edge and embedded systems where architecture and power are limited, traditional mixed-criticality systems (and storage) introduce a new dimension: accuracy. This dimension provides the greatest value in systems where the availability of data and sensor inputs exceeds the hardware’s capability to compute all potentially beneficial elements, and a processing-level imprecise calculation enables inputs to, in turn, inform accuracy needs. The key to achieving this at the hardware layer can be found in analog computing. For AI/ML systems, imprecise and analog processing approaches hold the potential to dramatically increase overall accuracy for a growing range of applications. An added benefit of analog computing is its resiliency to a variety of external interference – especially critical for lightweight space applications.