Energy scalable systems for 2D and 3D low-power ultrasound beamforming

Abstract

In traditional ultrasound imaging systems, bulky and power-intensive mainframes are used to process the high number of waveforms acquired in parallel from a large transducer array. The computational power of these systems scales linearly with transducer count. However, there exist applications where basic functionality in low-power conditions may be favorable to an "all-or-nothing" system that only produces a high resolution image when enough power is supplied. This thesis presents systems designed to support energy-scalability at run-time, enabling the user to make the tradeoff between power and performance. First, a system-level energy model for a receive-side digital beamforming system is presented. Power-performance tradeoffs for the analog front-end, analog-to-digital converter, and digital beamformer are analyzed individually and then combined to account for the performance dependency between the functional components. These considerations inform a recommendation on design choices for the end-to-end system. Second, this thesis describes an energy-scalable 2-D beamformer that provides user-controlled run-time tradeoff between image quality and energy consumption. Architectural design choices that enable three operating modes are discussed. A test chip was fabricated in 65-nm low power CMOS technology. It can operate with functional correctness at 0.49 V, with a measured power of 185 [mu]W in real-time operation at 0.52 V. Finally, a software-based energy-scalable 3-D ultrasound beamformer is implemented on an embedded supercomputer. The energy consumption and corresponding imaging quality are measured and compared.