Metal-oxide based, CMOS-compatible ECRAM for Deep Learning Accelerator

@article{Kim2019MetaloxideBC,
  title={Metal-oxide based, CMOS-compatible ECRAM for Deep Learning Accelerator},
  author={Seyoung Kim and J. A. Ott and Takashi Ando and H. Miyazoe and Vijay Narayanan and John Rozen and Teodor K. Todorov and Murat Onen and Tayfun Gokmen and Douglas M. Bishop and Paul M. Solomon and Ko-Tao Lee and Matthew Copel and Damon B. Farmer},
  journal={2019 IEEE International Electron Devices Meeting (IEDM)},
  year={2019},
  pages={35.7.1-35.7.4}
}
We demonstrate a CMOS-compatible, metal-oxide based Electro-Chemical Random-Access Memory (MO- ECRAM) for high-speed, low-power neuromorphic computing. The device demonstrates symmetric and linear conductance update, large on/off ratio and good retention while also withstanding high temperature treatments necessary for BEOL compatibility. Resistive switching in MO-ECRAM is observed with voltage pulses down to 10 ns and scales exponentially with voltage pulse amplitude, enabling parallel array… Expand

Figures from this paper

Transient Investigation of Metal-oxide based, CMOS-compatible ECRAM
Metal-oxide based Electrochemical Random-Access Memory (MO-ECRAM) has shown unique potential as a nonvolatile element for analog in-memory computation of deep learning tasks. Using a speciallyExpand
Experimental measurement of ungated channel region conductance in a multi-terminal, metal oxide-based ECRAM
  • Hyunjeong Kwak, Chuljun Lee, Chaeun Lee, K. Noh, Seyoung Kim
  • Physics
  • Semiconductor Science and Technology
  • 2021
Due to the rapid progress of artificial intelligence technology based on neural networks, the amount of required computation has been increasing dramatically. To keep up with the ever-increasingExpand
A back-end, CMOS compatible ferroelectric Field Effect Transistor for synaptic weights
TLDR
Voltage-controlled, symmetric and analog potentiation and depression of a ferroelectric Hf0.57Zr0.43O2 field effect transistor (FeFET) with good linearity is demonstrated, and the device concept has a great potential for the fabrication of high density, large-scale integrated arrays of artificial analog synapses. Expand
In situ Parallel Training of Analog Neural Network Using Electrochemical Random-Access Memory
TLDR
The implementation of the first in situ parallel training together with strong agreement with simulation results provides a significant advance toward developing ECRAM into larger crossbar arrays for artificial neural network accelerators, which could enable orders of magnitude improvements in energy efficiency of deep neural networks. Expand
CMOS-Compatible Protonic Programmable Resistor Based on Phosphosilicate Glass Electrolyte for Analog Deep Learning.
TLDR
This work reports on the first back-end CMOS-compatible nonvolatile protonic programmable resistor enabled by the integration of phosphosilicate glass as the proton solid electrolyte layer, which shows desirable modulation characteristics in terms of symmetry, retention, endurance, and energy efficiency. Expand
A BEOL Compatible, 2-Terminals, Ferroelectric Analog Non-Volatile Memory
A Ferroelectric Analog Non-Volatile Memory based on a WOxelectrode and ferroelectric HfZrO4 layer is fabricated at a low thermal budget (~375°C), enabling BEOL processes and CMOS integration. TheExpand
Algorithm for Training Neural Networks on Resistive Device Arrays
TLDR
A new training algorithm, so-called the “Tiki-Taka” algorithm, is presented that eliminates this stringent symmetry requirement for resistive crossbar arrays and maintains the aforementioned power and speed benefits. Expand
Emerging Materials for Neuromorphic Devices and Systems
TLDR
The functions of biological synapses and neurons are discussed, and deep neural networks and spiking neural networks are described, and the operation mechanism and the neuromorphic functions of emerging devices are reviewed. Expand
Edge AI without Compromise: Efficient, Versatile and Accurate Neurocomputing in Resistive Random-Access Memory
TLDR
NeuRRAM is presented, the first multimodal edge AI chip using RRAM CIM to simultaneously deliver a high degree of versatility in reconfiguring a single chip for diverse model architectures, record energy-efficiency better than prior art across various computational bit-precisions, and inference accuracy comparable to software models with 4-bit weights. Expand
Pr0.7Ca0.3MnO3-Based Three-Terminal Synapse for Neuromorphic Computing
In this letter, we report an all-solid-state Pr0.7 Ca0.3MnO3 (PCMO)-based three-terminal synapse device for neuromorphic computing. In this device structure, PCMO channel conductance can be preciselyExpand
...
1
2
...