This paper presents the design of a DFE for a 2x blind ADC-based RX. The DFE is implemented in 65-nm CMOS along with a 2x blind CDR and ADC. Our measured results confirm 5Gb/s data recovery with BER less than 10 -12 with a channel introducing 13.3dB of attenuation at the Nyquist frequency of 2.5GHz. Without the DFE, the BER exceeds 10 -8 . Keywords: speculative DFE, blind sampling, ADC-based receiver and CMOS Introduction An ADC-based receiver converts the incoming analog signal to its digital counterpart in order to enable significant equalization in the digital domain, especially at higher data rates where the channel loss is severe [1] [2]. The architecture proposed in [1] uses a blind clock (at twice the baud rate) for the ADC to sample the incoming signal and feeds these digital samples to a feed-forward equalizer (FFE) to remove the inter-symbol interference (ISI). An FFE, however, is known to suffer from noise enhancement and, as such, is inferior to a non-linear equalizer such as a decision-feedback equalizer (DFE). This paper presents for the first time a DFE for 2x blind ADC-based receivers. Fig. 1 compares a DFE in a phase-tracking receiver (RX) with a DFE in a blind RX. In a phase-tracking RX, the DFE relies on the recovered clock from the incoming signal for its operation [2]. In a blind ADC-based RX, however, there is no recovered clock [1]; the sampling clock is totally blind. This makes the task of DFE design nontrivial as the samples of the incoming signal no longer correspond to the centre of the eye or to any specific data phase. Moreover, the locations of the samples change during the operation of the receiver (e.g. due to low-frequency jitter) necessitating variable DFE coefficients.Proposed DFE Scheme Fig. 2 illustrates the proposed DFE scheme. Fig. 2 (left) shows the actual pulse response of the channel corresponding to UI n-1 ; the tail of this signal stretches over the next UI (UI n ) due to the limited bandwidth of the channel. The desired pulse response is derived such that the sum of two consecutive pulses results in an approximately constant value. The shaded area depicts the ISI; the amount of this ISI is a function of the sampling time. Therefore, as shown in Fig. 2 (right), the proposed DFE divides the nominal UI into 8 equal intervals, I [0:7] , and assigns one coefficient to each interval. These coefficients, α [0:7] , represent the interference from b n-1 in I [0:7] . During the operation of the RX, the DFE uses the average transition phase, Φ AVG , recovered by the clock and data recovery (CDR) [1] to select two coefficients among 8, which correspond to the sampling time. Φ AVG1 , the modulo-1UI version of Φ AVG , indicates the distance between the second sample of the current UI, S 2 , and the next nominal UI boundary. As an example, Fig. 2 (right) illustrates the case where S 2 falls in I 2 , hence α 6 and α 2 will be used for S 1 and S 2 , respectively. Fig. 3 depicts the full-rate implementation of the proposed RX in a simplified block diagram. Two ADC samples o...