2022 IEEE Symposium on VLSI Technology and Circuits (VLSI Technology and Circuits) · 2022 · 13 citations · 0 references
EngineeringComputer EngineeringElectrical EngineeringAnalog ElectronicsLow-power DesignSemiconductor DevicesDynamic Power ReductionEmbedded Memory BlocksMemory DevicesMemory Subsystem DesignMemory CircuitsMemory Interface ArchitectureMemory Interface DesignOffset CalibrationPower ConsumptionPower EfficiencyPower-efficient Memory DesignSelf-training
This paper presents a 375-GB/s/mm power-efficient memory interface that consists of the PAM-4 transceivers with a per-pin training system for the next-generation HBM controllers. The self-training system executes foreground driver calibration, 2-D sampling point optimization, FFE coefficient adaptation, and sampler offset calibration. Using DC-levels and SBR patterns, the entire training sequence for 8 DQ transceivers and 2 DQS transceivers takes less than 1-ms. In addition, a charge-recycling sampler that saves 44.5% of power consumption compared to the strongARM latch sampler is proposed. The proposed memory interface fabricated in a 40-nm CMOS process shows a power efficiency of 0.41-pJ/b and a power efficiency per channel length of 68.7-fJ/b/mm, significantly higher than the standard state-of-the-art memory interfaces.