The goal of this research is to develop high-performance and power-efficient digital signal processors, using a novel event-driven approach involving special techniques that carefully keep track of the timing of events. Unlike standard "asynchronous" logic this approach inherently uses physical time as part of its representation of information. The systems and techniques to be developed in course of this research can benefit a number of important applications, including sensor networks and biomedical instrumentation in health care. The body of knowledge to be created would potentially find applications in other areas as well, including control systems and robotics. The work has a significant educational component, in that it involves synergistic development, combining different domains of knowledge and thus training the graduate and undergraduate students involved in interdisciplinary work, leading to new, creative solutions.
Appealing technical properties of this approach include the absence of aliasing, better spectral properties of error, and faster response to changing inputs, compared to classical techniques. The approach is suitable for certain applications that demand very low power dissipation, notably those involving portability or difficult-to-access locations. While prior work concentrated on demonstrating the feasibility of continuous-time digital signal processors, this new effort concentrates on developing next-generation systems and achieving very high performance, necessitating new principles and techniques. The work seeks to advance the performance of continuous-time digital signal processors by reducing the effective sampling rate and improving output reconstruction, by using special coding techniques for relaxing granularity requirements in the signal processing units, by processing directly non-uniformly quantized signals, and by employing algorithmic techniques; it also expands continuous-time digital techniques to general scientific computation. The proposed effort brings together two groups with complementary expertise. The Columbia University group specializes in mixed-signal circuits and has demonstrated the first continuous-time digital signal processors; the Cornell University group works in asynchronous digital electronics, including event-driven processors and asynchronous dataflow FPGAs.
