The scaling of CMOS technology continues to improve the processor capability and memory capacity, requiring memory interface with higher bandwidth and better energy efficiency to enhance the overall system performance. Among those cutting-edge designs of memory interface, multi-band signaling has shown great potential for its high throughput along with low energy consumption. With spectrally divided signaling, the multi-band transceiver can be designed to avoid spectral notches and extend communication bandwidth on multi-drop buses. Also, the multi-band transceiver is immune to inter-symbol interference caused by channel attenuation because of unique self-equalization double-sideband signaling. In this dissertation, we will demonstrate a tri-band transceiver with four parallel lanes that achieved a total data rate of 40Gb/s with total power consumption of 38mW in 28nm CMOS technology.
To realize the total data rate of 40Gb/s, PAM-4 and 16-QAM are used at the baseband and 3/6GHz bands, respectively, to carry 10 parallel bit streams at 1GBaud via each lane of the transceivers. These ten parallel bit streams share the same physical channel to minimize the time skew among them. In view of this, the strobe signal, DQS, is assigned to one of the ten bits for data recovery at the receiving end. Under 6dB attenuation at 6GHz on a 2” dense FR-4 differential bus (line pitch of 6mil), the transmitting end consumes only 1.6mW/lane. Together with 4.7mW/lane of the receiving end and 13.4mW of the carrier generator to be shared among all lanes consumes, the total power consumption and average energy efficiency are 38mW and 0.95pJ/b. Compared with prior arts, the proposed design achieves not only better energy efficiency but also substantial size advantage (0.01mm2/lane, including the carrier generator). This transceiver realizes a total data rate of 40Gb/s with BER < 10^-12. Moreover, this tri-band architecture can be scaled in the frequency domain for further increasing the data throughput without increasing the symbol rate, which enables a new design dimension with more compact size and significantly improved energy efficiency for future memory interfaces.
