A 20Gb/s Forwarded Clock Transceiver in 90nm CMOS B.

Future microprocessor platforms will require system-level optimization of the I/O to minimize cost and maximize aggregate bandwidth. Critical parameters such as silicon area, power, testability, and off-chip interconnect quality must be properly balanced to maximize the I/O performance versus cost ratio. For example, there is a fundamental tradeoff between clock quality and equalizer effectiveness 1. Producing precision RX and TX clocks and a sensitive RX may impact the power and area of some circuits, but it could allow the use of simple, low-power linear equalizers to minimize the overall link power. Additionally, these equalizers will become much more effective by limiting near-end crosstalk and stubbed backplane (BP) via length. To demonstrate this system-level optimization effort, we have developed a 20Gb/s forwarded clock I/O system intended for a wide parallel link with small area and low power.