In memory interfaces, data rates continue to increase. However, channel loss attenuates and distorts high-speed data, which degrades signal integrity. To address these issues, pulse-amplitude modulation (PAM) has been adopted in memory interfaces as a...
In memory interfaces, data rates continue to increase. However, channel loss attenuates and distorts high-speed data, which degrades signal integrity. To address these issues, pulse-amplitude modulation (PAM) has been adopted in memory interfaces as an alternative to non-return-to-zero (NRZ) signaling. Three-level pulseamplitude modulation (PAM-3) provides 1.5 times higher bit efficiency than NRZ and 1.5 times larger voltage margin than four-level pulse-amplitude modulation (PAM-4), offering a balanced trade-off between bandwidth efficiency and voltagemargin. However, PAM signaling still suffers from reduced voltage margin and increased circuit complexity, which increases power consumption and makes it more susceptible to inter-symbol interference (ISI). Therefore, advanced transceiver (TRX) architectures are required to ensure high-speed and low-power operation. This paper proposes transmitter (TX) and receiver (RX) architectures for a single-ended PAM-3 transceiver. Conventional feed-forward equalizer (FFE) designs increase the load capacitance of the output driver, which degrades bandwidth and energy efficiency. To address this, the proposed TX adopts a switched-capacitor peaking (SCP) TX that minimizes driver loading while simplifying the equalizer structure. Conventional single-ended PAM-3 receivers require single-to-differential converters (S2Ds) and reference voltage generators, which increase power consumption and circuit complexity. The proposed RX employs a summer-merged transimpedance amplifier (SMTIA) that merges decision-feedback equalization (DFE) into the TIA, thereby removing the need for S2D and reference generation. As a result, both power consumption and area overhead are reduced. In addition, an active inductor provides high-frequency boosting, effectively mitigating ISI caused by channel loss. Simulation results of the proposed TX, implemented in a 28-nm CMOS process, show -9.2 dB channel loss compensation at 8 GHz with an energy efficiency of 0.127 pJ/b/dB, achieving a vertical eye margin of 61 mV and a horizontal margin of 0.65 UI. The proposed RX achieves 24 Gb/s operation under -6.55 dB channel attenuation with an energy efficiency of 0.39 pJ/b, providing a vertical eye margin of 204 mV and a horizontal margin of 0.69 UI.