For low-power 8.5-Gbps operation, 4th-generation 10-nm 16-Gb LPDDR5x DRAM I/O circuits and control methods are proposed in this article. The proposed I/O improves signal integrity (SI) and power integrity (PI) by using a self-pre-emphasized stacked driver in transmitter (Tx), a supply voltage insensitive data receiver (Rx), and an optimized clock tree in write clock (WCK) clock paths. The measured eye widths of Tx and Rx, which are the indicators of the SI/PI, are 0.66 unit interval (UI) and 0.57 UI at 8.5 Gbps, respectively. Also, the measured power consumption is reduced by 20.0% compared to the previous LPDDR5 SDRAM product.
A 4Gb/s/pin 32b parallel 512Mb GDDR4 SDRAM is implemented in an 80nm DRAM process. It employs a data-bus inversion coding scheme with an analog majority voter insensitive to mismatch, which reduces peak-to-peak jitter by 21 ps and voltage fluctuation by 68mV. A dual duty-cycle corrector is proposed to average duty error, and tuning is added to the auto-calibration of driver and termination impedance.
This paper describes pseudo-differential signaling schemes for DRAM interfaces that minimize the skew between data and suppress common-mode noise. The chip is implemented in a 0.13 mum process and occupies 2.7times2.3 mm2 and the active area is 1.0times0.3 mm2.
15 January 2019 in IEEE Journal of Solid-State Circuits
This Special Issue of the IEEE Journal of Solid-State Circuits is dedicated to a set of the best papers selected from the 2018 IEEE International Solid-State Circuits Conference (ISSCC) that took place in San Francisco, CA, USA, on February 11–15, 2018. It covers papers from the Wireline (WLN), Digital Circuits (DCT), Digital Architectures and Systems (DAS), and Memory (MEM) committees.
With the explosive data growth due to HD video content, IoT, and the rapid shift from enterprise to cloud computing, Mega Data Centers have become an essential part of the global network infrastructure. Despite their massive scale, Data Center architectures require software-defined network capability and scalability, with lower latency. Interconnects will need to support diversified media-types, higher data-rates (>50Gb/s per lane), and longer distance (up to 2km) to achieve this and still make economic sense. Once a niche for mainly VCSEL links in HPC, “Optical Ethernet” is now becoming mainstream as interconnects for both MMF and SMF in various fiber configurations are being defined within the standards. Also, copper links including backplane continue to be the interconnect with highest volume. The Forum will cover various wireline transceivers enabling Mega Data Centers such as ADC-based transceivers and Silicon Photonics, and challenges moving forward.
This paper addresses the dual challenge of improving anomaly detection and signal integrity in high-speed dynamic random access memory signals. To achieve this, we propose a joint training framework that integrates an autoencoder with a classifier to learn more distinctive latent representations by focusing on valid data features. Our approach is evaluated across three anomaly detection algorithms and consistently outperforms two baseline methods. Detailed ablation studies further support these findings. Furthermore, we introduce a signal integrity enhancement algorithm that improves signal integrity by an average of 11.3%. The source code and data used in this study are available at https://github.com/Usama1002/learning-latent-representations.