A single-ended four-level pulse-amplitude modulation (PAM-4) transmitter (TX) for memory interfaces achieves high signal integrity by combining an impedance-matched PAM-4 driver with a three-point ZQ calibration scheme. This improves PAM-4 linearity by allowing the driver to compensate for its impedance variation caused by the change in the drain-source voltage (VDS) to suit the four output levels considering both the TX and the receiver (RX). Resistors and inductors are eliminated from the voltage-mode (VM) driver, reducing the area requirement. The two-tap asymmetric feed-forward equalization (FFE) allocates six different coefficients to each minimum pull-up and pull-down transition, compensating for nonlinear equalization strengths and asymmetric characteristics of the driver. A prototype chip fabricated in the 65-nm CMOS has an area of 0.0333 mm² and consumes 0.64 pJ/bit. It achieves a data rate of 28 Gb/s/pin with a ratio level separation mismatch (RLM) of 0.993.

Currently, we are living in a data-centric era as the need for large amounts of data has dramatically increased due to the widespread adoption of artificial intelligence (AI) in a variety of technology domains. In the current computing architecture, the memory input and output (I/O) bandwidth is becoming a bottleneck for improving computing performance; therefore, high-bandwidth memory interfaces are essential. In addition, the high power consumption of data centers to edge AI devices will lead to power shortages and climate crises in the near future; therefore, energy-efficient techniques for memory interfaces are also important. This paper presents contemporary approaches to improve I/O bandwidth, such as increasing the I/O pin count and data rate/pin, and to save energy in memory interfaces. However, there are still some design challenges that require further improvements. Therefore, various design challenges and problems to be solved are discussed, and future perspectives, including chiplet and die-to-die interfaces, are presented. Based on various research and development efforts to overcome the current limitations, the technological paradigm shift and related industries are expected to advance to the next level.

A quadrature clock corrector uses relaxation oscillators to detect duty-cycle and quadrature phase errors by transforming them into pairs of frequencies, which are then digitized and compared. It achieves good detection accuracy and can detect a wide range of duty-cycle and quadrature phase errors. The prototype is implemented in a 55-nm CMOS process with a supply voltage of 1.2 V and occupies an area of 0.003 mm ² . The experimental results show that the operation range is from 1 to 3 GHz, the power efficiency is 0.79 mW/GHz, the maximum duty-cycle error is 0.8% at 3 GHz, and the maximum quadrature phase error is 1.1° at 3 GHz.

The demand for high-performance graphics systems used for artificial intelligence, cloud game, and virtual reality continues to grow; this trend requires graphics systems to achieve ever higher bandwidths. This article proposes a GDDR6 dynamic random access memory (DRAM) with a half-rate clocking architecture and optimized receiver and transmitter to improve high-speed operation. Furthermore, this article adopts a staggered PAD using the redistribution layer (RDL) to reduce the distance to four PADs; it enables the mitigation of bandwidth limitation of half-rate clocking, a lower phase mismatch, and a reduced propagation delay. The proposed half-rate clocking-based GDDR6 DRAM achieves 24 Gb/s/pin on a 1.35-V DRAM process. Also, the power-supply-induced-jitter (PSIJ) value is improved from 9.97 to 3.22 ps, compared to a GDDR6 design using a quarter-rate clocking. In addition, the phase mismatch of the proposed clock distribution network (CDN) is reduced compared to the conventional CDN, resulting in an improvement of the 3-σ value of the phase skew from 4.16 to 2.25 ps.

We combine 2-tap feed-forward amplitude equalization with phase equalization by 4-tap integrated pulse-width modulation. In a VSS-terminated transmitter, amplitude equalization is selected for pull-up data transmission, and phase equalization for pull-down data transmission, and the strength of equalization can be controlled depending on channel losses. This combines the strength of amplitude equalization with the energy efficiency of phase equalization. A prototype quarter-rate transmitter for memory interfaces, fabricated in a 65nm CMOS process, performed single-ended signaling at a data-rate of 16Gb/s/pin over a channel with a loss of 14.7dB. Its energy efficiency is 1.04pJ/bit/pin and the figure-of-merit is 0.070pJ/bit/pin/dB.

In asymmetric memory systems, the pull-up and pull-down data from the channel can have different amounts of inter-symbol interference (ISI), so they cannot be fully corrected by an equalizer which uses a single tap weight. We introduce a receiver-side single-ended 1-tap asymmetric decision-feedback equalizer (DFE) with a weight selection multiplexer, which allows the application of a different weight to each direction of data value transition. Implemented in a 55nm CMOS technology, our DFE compensated for asymmetric ISI in a 10.4Gb/s signal from a channel with an insertion loss of -8.3dB, leading to a 50% wider horizontal margin at a bit-error-rate of 10-12 than symmetric correction. The energy efficiency is 0.16pJ/bit.