Bulk-Switching Memristor-Based Compute-In-Memory Module for Deep Neural Network Training.

The constant drive to achieve higher performance in deep neural networks (DNNs) has led to the proliferation of very large models. Model training, however, requires intensive computation time and energy. Memristor-based compute-in-memory (CIM) modules can perform vector-matrix multiplication (VMM) in place and in parallel, and have shown great promises in DNN inference applications. However, CIM-based model training faces challenges due to non-linear weight updates, device variations, and low-precision. In this work, a mixed-precision training scheme is experimentally implemented to mitigate these effects using a bulk-switching memristor-based CIM module. Low-precision CIM modules are used to accelerate the expensive VMM operations, with high-precision weight updates accumulated in digital units. Memristor devices are only changed when the accumulated weight update value exceeds a pre-defined threshold. The proposed scheme is implemented with a system-onchip of fully integrated analog CIM modules and digital sub-systems, showing fast convergence of LeNet training to 97.73%. The efficacy of training larger models is evaluated using realistic hardware parameters and verifies that CIM modules can enable efficient mix-precision DNN training with accuracy comparable to full-precision software-trained models. Additionally, models trained on chip are inherently robust to hardware variations, allowing direct mapping to CIM inference chips without additional re-training.

Synthesis and characterization of volatile, fluorine-free beta-ketoiminate lanthanide MOCVD precursors and their implementation in low-temperature growth of epitaxial CeO(2) buffer layers for superconducting electronics.

A new class of volatile, low-melting, fluorine-free lanthanide metal-organic chemical vapor deposition (MOCVD) precursors has been developed. The neutral, monomeric Ce, Nd, Gd, and Er complexes are coordinatively saturated by a versatile, multidentate ether-functionalized beta-ketoiminato ligand series, the melting point and volatility characteristics of which can be tuned by altering the alkyl substituents on the keto, imino, and ether sites of the ligand. Direct comparison with conventional lanthanide beta-diketonate complexes reveals that the present precursor class is a superior choice for lanthanide oxide MOCVD. Epitaxial CeO(2) buffer layer films can be grown on (001) YSZ substrates by MOCVD at significantly lower temperatures (450-650 degrees C) than previously possible by using one of the newly developed cerium beta-ketoiminate precursors. Films deposited at 540 degrees C have good out-of-plane (Deltaomega = 0.85 degrees ) and in-plane (Deltaphi = 1.65 degrees ) alignment and smooth surfaces (rms roughness approximately 4.3 A). The film growth rate decreases and the films tend to be smoother as the deposition temperature is increased. High-quality yttrium barium copper oxide (YBCO) films grown on these CeO(2) buffer layers by pulsed organometallic molecular beam epitaxy exhibit very good electrical transport properties (T(c) = 86.5 K, J(c) = 1.08 x 10(6) A/cm(2) at 77.4 K).