Micronets: a model for decentralising control in asynchronous processor architectures

Micronets model processor architectures as a network of communicating resources, in contrast to the traditional one of a linear pipeline. Micronets distribute the control to the functional units, which enables the exploitation of ne-grain concurrency between instructions. The overhead due to asynchrony is hidden with the four-phase protocol being used to implement scoreboarding and hazard avoidance mechanisms, without incurring additional control costs. This paper demonstrates the feasibility of micronet-based processors. Results are presented for SPICE-level simulations of a 0.7m CMOS implementation of a datapath. The relationships between micronets and both the compiler and the computer architecture are also explored.