Tin Tin Wee

A Versatile 90-nm CMOS Charge-Pump PLL for SerDes Transmitter Clocking

This paper presents a low-jitter charge-pump phase-locked loop (PLL) built in standard 90-nm CMOS for 1 to 10 Gb/s wireline SerDes transmitter clocking. The PLL employs a programmable dual-path loop filter with integral path and resistorless sample-reset proportional path that are independently controlled for flexible setting of closed-loop bandwidth and peaking. Frequency is synthesized by a digitally calibrated LC-VCO achieving 45% calibration tuning range with inversion-mode nMOS varactors and area-efficient helical inductors. Following calibration, 4.8% hold range compensates for VCO sensitivity to supply voltage and temperature drift. The PLL exhibits 0.81 ps rms jitter at 10 Gb/s. Critical for ASICs integrating noisy digital cores and multiple SerDes channels, design considerations to minimize jitter induced by supply noise are described. Deep-submicron CMOS effects on design are also examined to improve manufacturability and performance

An 8.0-Gb/s HyperTransport Transceiver for 32-nm SOI-CMOS Server Processors

We present an 8.0-Gb/s HyperTransport source-synchronous I/O integrated in a 32-nm SOI-CMOS processor for high-performance servers. Based on a 45-nm design capping at 6.4 Gb/s, the 32-nm transceiver achieves up to 8.0 Gb/s over long-reach board channels by incorporating several jitter- and power-reduction enhancements. First, a high-bandwidth digital clean-up PLL is introduced to attenuate high-frequency jitter in the received forwarded clock before the data is sampled. This PLL achieves a highly programmable jitter bandwidth of 20-296 MHz (measured with 0.2 UIpp input jitter) and 0.90-1.50 ps output rms jitter by implementing an extended bang-bang phase detector for additional phase-error magnitude information and flexible bang-bang control of a current-starved ring-based oscillator. Second, several power-hungry circuits, namely the transmitter input FIFO and output driver as well as the receiver deserializer, are redesigned for 8.0-Gb/s operation to maintain thermal compatibility with the existing 45-nm socket package. The fabricated 20-lane I/O consumes 1.70 W at 8.0 Gb/s with an energy efficiency of 11.8 pJ/bit. This reflects a 4.9% increase in HyperTransport power consumption and only 0.3% increase in total processor target power relative to 45-nm operation at 6.4 Gb/s.

Loopback architecture for wafer-level at-speed testing of embedded HyperTransport TM processor links

We present transceiver serial loopback that enables cost-effective wafer-level at-speed testing of HyperTransportTM (HT) I/O for processor die-to-die communication. Besides facilitating known-good-die testing, this feature provides observability of multi-chip module (MCM) die-to-die links that are completely embedded without external pin visibility. We demonstrate production screening of 45-nm SOI-CMOS wafers at 6.4 Gb/s.

Extending HyperTransport™ technology to 8.0 Gb/s in 32-nm SOI-CMOS processors

We present an 8.0-Gb/s HyperTransport™ technology I/O built in a 32-nm SOI-CMOS processor for high-performance servers. Based on a 45-nm design that caps at 6.4 Gb/s, the 32-nm transceiver achieves up to 8.0 Gb/s over long-reach board channels. Key enhancements include a high-bandwidth (>200 MHz) PLL to attenuate high-frequency jitter in the received forwarded clock and redesigned power-hungry circuits to operate at 8.0 Gb/s within the existing 45-nm package thermal limit.

Denver Hosts Technical Seminars on Cutting-Edge CMOS Technology and High-Speed Test

In the past year, the Denver SSCS Chapter hosted eight monthly seminars, including four by SSCS Distinguished Lecturers. These talks spanned a variety of exciting developments in IC design, cutting-edge CMOS technology, and high-speed test.