For wireline transmitters delivering a high-speed multi-level signal, such as pulse-amplitude-modulation-4 or duobinary, a high-performance multiplexer (MUX) is critical to serialize the low-speed parallel data into one full-speed output. To enhance the power efficiency and data eye's opening, this paper proposes a universal 2-to-1 MUX, featuring a cross-quadrature clocking technique to enlarge the timing margin, and a simplified three-latch topology without delay buffers to boost the internal bandwidth (BW). The MUX ratios are extendable to 4-to-2 and 4-to-1, and their benefits are exemplified via a duobinary-signal transmitter. It further includes an output driver unifying the MUX-and-SUM operation, a BW-extended single-to-differential converter, and an active-inductor-embedded clock buffer for swing enhancement. Also, a predictive method for estimating the duobinary-signal data-dependent jitter according to the load capacitance of the output driver is developed. Fabricated in 65-nm CMOS, the transmitter exhibits a figure-of-merit of 1.3 mW/Gb/s at 36 Gb/s, while occupying a compact die area of 0.037 mm.