30.4 A 32Gb/s 2.9pJ/b Transceiver for Sequence-Coded PAM-4 Signalling with 4-to-6dB SNR Gain in 28nm FDSOI CMOS

Multilevel signaling such as PAM-4 makes more efficient use of the better part of the channel response by mapping multiple bits in the same time interval (i.e. UI). At the same time, to achieve lower power consumption simple equalization techniques such as TX-FIR and RX peaking equalization are often preferred. For example, in Fig. 30.4.1, a channel with 30dB of loss is equalized using these linear equalization techniques. However, the increased spectrum efficiency comes at the cost of SNR that is captured in the single bit response (SBR). Here, the equalized symbol response can be decomposed into MSB and LSB SBRs. In PAM-4 signaling, individual eye heights are set by the LSB SBR, whereas, the total signal amplitude is set by the MSB and LSB combined SBR. This 1/3 ratio suggests a 9dB SNR penalty, but in reality, crosstalk and residual ISI noise also get enhanced. Since the residual ISI is proportional to the pulse amplitude, the LSB will experience both MSB and LSB residual ISI that is also 3× worse compared to NRZ, and the same is also true for crosstalk. Obviously, this reduction in signalto-noise-and-crosstalk ratio is compensated by improved channel characteristics to make PAM-4 signaling justified and advantageous over NRZ. However, as the channel loss increases beyond 20dB, the peak power constraint of the transmitter causes further SNR loss. Even with high frequency boost and gain provided by the linear equalizer, eye opening is less than 30mV at the sampler input and this limits the achievable symbol error rate in the link [1].