Skip to Main Content
This paper presents a low-power decision-feedback equalizer (DFE) receiver front-end and a two-step minimum bit-error-rate (BER) adaptation algorithm. A high energy efficiency of 0.46 mW/Gbps is made possible by the combination of a direct-feedback finite-impulse-response (FIR) DFE, an infinite-impulse-response (IIR) DFE, and a clock-and-data recovery (CDR) circuit with adjustable timing offsets. Based on this architecture, the power-hungry stages used in prior DFE receivers such as the continuous-time linear equalizer (CTLE), the current-mode summing circuit for a multitap DFE, and the fast selection logic for a loop-unrolling DFE can all be removed. A two-step adaptation algorithm that finds the equalizer coefficients minimizing the BER is described. First, an extra data sampler with adjustable voltage and timing offsets measures the single-bit response (SBR) of the channel and coarsely tunes the initial coefficient values in the foreground. Next, the same circuit measures the eye-opening and bit-error rates and fine tunes the coefficients in background using a stochastic hill-climbing algorithm. A prototype DFE receiver fabricated in a 65-nm LP/RF CMOS dissipates 2.3 mW and demonstrates measured eye-opening values of 174 mV pp and 0.66 UIpp while operating at 5 Gb/s with a -15-dB loss channel.