Adaptive data-transition decision feedback equalizer for serial links

Data-state decision feedback equalizers (DFEs) suffer from a fundamental drawback of deteriorating vertical eye-opening when consecutive 1s or 0s are present in data. To combat this, a new data-transition adaptive DFE termed data-transition DFE is proposed. We show that data-transition DFE does not reduce vertical eye-opening whereas data-state DFE shrinks vertical eye-opening when consecutive 1s or 0s are present. We further show for the high-frequency components of data, datatransition DFE is capable of increasing vertical eye-opening. Although data-state DFE is also capable of increasing vertical eye-opening for the high-frequency components of data, this is at the expense of sacrificing vertical eye-opening for the low-frequency components of data. The stronger the DFE action, the severer the reduction of the vertical eye-opening of the low-frequency components of data. The optimal tap of data-state DFE occurs when the vertical eye-opening of the low-frequency components of data is the same as that of the high-frequency components of data. Moreover, we show that data-transition DFE not only offers a unity signal transfer function at low frequencies where most of the energy of data is located but also provides first-order shaping on the difference between the desired and equalized data. Both give rise large vertical eye-opening. The theoretical findings are validated using the simulation results of two serial links, one with data-state DFE with loop-unrolling and the other with the data-transition DFE designed in TSMC 65 nm CMOS technology.