Real-time VLSI architecture for bio-medical monitoring

This paper discusses the architecture and implementation of SSS2, a high-performance real-time signal processing system developed with a hybrid ESL/RTL methodology and targeted to biomedical image processing. Traditional methodologies, as well as new tools, such as Cebatech's C2R untimed-C synthesizer have been employed in the design of the system. The SSS2 platform specifies a parametric number of scalar processing elements, based on multiple 32-bit Sparc-compliant engines, augmented with LE2, an ESL-designed 2-way LIW/SIMD accelerator. LE2, which is purely designed in C, exposes a consistent interface to its SIMD datapath directly which is directly derived from the C-source of open-source image processing codes. It is synthesized to Verilog RTL with C2R. Behaviorally-synthesized SIMD datapaths are then 'plugged-in' into the exposed LE2 datapath interface. The LE2 memory interface can be either a cache- based configurable vector load/store unit or a multi-banked, multi-channel streaming local memory system. Results drawn from this work strongly suggest a shift towards a hybrid approach in designing multi-core systems for high bandwidth streaming and for dealing with large scale medical image transfers and non-linear bio-signal processing algorithms.