A High-Throughput Asynchronous Multi-resource Arbiter Using a Pipelined Assignment Approach

Multi-resource arbitration is used where interchangeable resources need to be assigned among a number of client users. An efficient arbiter is thus critical to manage these shared resources. There has been only limited prior research on efficient asynchronous solutions, which have mainly focused on serial approaches. In this paper, a new high-throughput multi-resource arbiter is proposed, based on fine-grain pipelining. The allocation of a resource to a client is divided into several steps, where multiple successive client-resource pairs can be selected rapidly in sequence, and the completion of the assignments can overlap in parallel. In addition, a new static four-phase asynchronous pipeline is introduced and applied to the arbiter design, incorporating a highly-concurrent handshaking protocol. Experimental results were obtained for the proposed arbiters on two different sizes, 4×3 and 8×4. The new arbiters were implemented in a 45nm technology using a standard cell library. Pre-layout results show the proposed designs have 23.5% and 36.2% lower cycle time, respectively, for back-to-back assignment completion, compared to the baseline non-pipelined approach. In addition, in spite of the pipelining, the new arbiters exhibit slightly better latency in light traffic.