A 1.8 pJ/bit 16 × 16Gb/s source-synchronous parallel interface in 32nm SOI CMOS with receiver redundancy for link recalibration

Abstract

A source-synchronous I/O architecture is reported that includes redundant receiver lanes to enable lane recalibration with reduced power and area overhead. Key features and considerations of the proposed architecture are described. A proof-of-concept 16 lane, 16 Gb/s per lane source-synchronous I/O test chip was designed and fabricated in a 32 nm SOI CMOS technology. Several circuit techniques employed in the design of this test chip are described. These include a phase rotator based on current-integrating phase interpolator cores with architecture and circuit improvements to performance as compared to prior art, an active-inductor-based RX CTLE, and an 8:1 TX serializer with 8-phase clocking. Measurements demonstrate the operation of the test chip over ultra-short-reach channels with up to 10 dB of loss with greater than 30% timing margin. The I/O circuitry operates from 1 V supplies and achieves a power efficiency of better than 2 pJ/bit, making the proposed architecture suitable for use in high-density interconnect applications required for high-performance computing systems.