CryoMem: A 4K-300K 1.3GHz eDRAM Macro with Hybrid 2T-Gain-Cell in a 28nm Logic Process for Cryogenic Applications
Rakshith Saligram, Suman Datta, Arijit Raychowdhury
Abstract
In the pursuit of higher digital performance, as well as for finding new applications in emerging computing models and operating environments, low-temperature (300K to 150K) and cryogenic (<; 150K) computing is gaining momentum. In particular, space electronics (4K-200K), digital-control in fuel-cell electric vehicles (20K-80K) and digital-assist/ peripherals of quantum/ superconducting computers (10mK - 100K) require temperature-scalable process technologies and digital logic/memory [1-7]. Further, cryo-CMOS computing (100K - 150K) has been recently shown to be a significant booster for high-performance computing (HPC), with process retargeted for cryo-HPC [1]. For all these applications, there is a demand for high-density, large-capacity, high-bandwidth (BW) memory, which cannot be addressed by non-CMOS technologies (e.g., Vortex Transitional devices, Josephson Junction based Memory arrays etc. [2-7]) that suffer from single-temperature operation, low-density, poor-scalability, poor-reliability and high design-complexity. On the other hand, scaled CMOS with improved characteristics at low-temperature, such as steep sub-threshold slope (SS), improved channel transport, low- I <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">OFF</sub> and reduced thermal noise, provides a promising, yet largely unexplored pathway for integrating large on-die memory with both CMOS and non-CMOS cryo-computing, across a wide range of temperatures and applications. In this research test-chip, we present a 2T-gain-cell (GC) based embedded-DRAM (eDRAM) macro in 28nm HKMG CMOS targeted for a range of cryo-applications and enabled by superior transistor characteristics at low-temperature. It features: (1) a hybrid P/N gain-cell for stable storage and low coupling noise, (2) an open bit-line architecture taking advantage of the low noise, (3) reliable operation from 300K to 4K and (3) 10 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">6</sup> × improvement of retention time from 300K to 4K. The cell architecture, operating voltages and the design-space for eDRAM (1T-1C and gain-cell) are summarized in Fig. 1. While low storage capacitor in eDRAM on logic leads to low retention time and high refresh power, the ultra-low leakage at cryo-temperature makes it a promising technology with a measured retention time of >1s at 4K.