UCLA

In this work, phase-based methods for emerging applications implemented with advanced CMOS technology are proposed. We use phase-based methods to solve problems in high- speed chip-chip/chip-memory data link interconnect, and in terahertz system.

With increasing demand for bandwidth between chip-chip and chip-memory communication, high-speed data link interconnect becomes a key block in modern computer systems. In or- der to serve portable devices, energy efficiency is a critical metric for data link interconnect. Multiband RF Interconnect, which transmits data by upconverts data streams into multiple frequency bands, can fully exploit the bandwidth of interconnect wire with excellent energy efficiency. In order to demodulate data stream, a correct carrier phase in receiver end is necessary. The conventional way for carrier phase synchronization requires high-speed, high- resolution analog-to-digital converter, which consumes large amount of power and silicon area. In the proposed phase-based method, a pilot signal is sent by transmit side, and in receiver end a simple comparator is sufficient to identify correct phase. This greatly enhances energy efficiency of data link and reduces cost. In addition, optimal demodulation phase for dispersive channel is derived.

The other proposed application for phase-based method is terahertz systems. CMOS tera- hertz source suffer from low device cut-off frequency, and fundamental frequency is limited.

Conventional time-domain harmonic extraction can selectively enhance certain harmonic of fundamental tone at output radiation and suppress other harmonics, and therefore increase effective radiation frequency. In terahertz band, time-domain harmonic extraction cannot achieve sufficient low-order harmonic suppression. A spatial-domain harmonic extraction method is proposed, which samples signal in specific spatial positions and combines sampled signals together to selectively enhance and eliminate tones. With proposed method, 1.4-THz CMOS imaging is achieved. In addition, a fast characterization method for terahertz source is proposed. This method uses random sampling (compressive sensing), and reconstruct sparse spectral profile by two-step zoom-in algorithm. With proposed method, a 8X time re- duction can be achieved for terahertz source characterization, which enables regular self-test of terahertz systems.