Periodic Reporting for period 1 - PHASTRAC (Phase Change Materials for Energy Efficient Edge Computing)
Periodo di rendicontazione: 2023-01-01 al 2024-06-30
In recent years, we have witnessed an explosion of artificial intelligence (AI) applications which will continue to grow over the next decade. An intelligent and digitized society will be ubiquitous, enabled by increased advances in nanoelectronics. Key drivers will be sensors interfacing with the physical world and taking appropriate action promptly while operating with energy efficiency and flexibility to adapt. Most sensors receive analog inputs from the real world and generate analog signals to be processed. However, digitizing these signals creates not only an enormous amount of raw data but also requires a lot of memory and high-energy consumption. As the number of sensor-based IoTs grows, bandwidth limitations and privacy aspects make it impossible to send everything back to a cloud for real-time processing and decision-making, especially for delay or privacy-sensitive applications such as driverless vehicles, robotics, personal healthcare, or industrial manufacturing.
In this context, PHASTRAC aims to develop a novel analog-to-information neuromorphic computing paradigm based on oscillatory neural networks (ONNs). We offer a first-of-its-kind and novel analog ONN computing architecture to seamlessly interface with sensors and process their analog data without any analog-to-digital conversion. ONNs are a biologically inspired neuromorphic computing architecture, where neuron oscillatory behavior will be developed by innovative phase change VO2 material coupled with synapses developed by bilayer Mo/HfO2 RRAM devices. PHASTRAC will address the most critical issues, namely 1) novel devices for implementing ONN architecture, 2) novel ONN architecture to allow analog sensor data processing, and 3) processing the data efficiently to take appropriate action. This “sensing-to-action” computing approach based on ONN technology will allow energy efficiency improvements of 100x-1000x and establish a novel analog computing paradigm for improved human-machine interactions.
The main objectives of PHASTRAC are:
(1) fabrication of low-power devices based on phase-change materials and bilayer metal-oxide RRAM devices that can serve as the building blocks for oscillatory neural networks (ONNs)
(2) development of devices, circuits to architecture-level models to allow exploration of the oscillatory neural network computing paradigm to allow the development of dedicated algorithms for ONN training
(3) development of a sensing-to-action computing paradigm with ONN that can serve as an energy efficient computing platform with potential interests in edge computing
(4) development of demonstrators and applications to show the advantage of ONN computing for AI tasks
In this context, PHASTRAC aims to develop a novel analog-to-information neuromorphic computing paradigm based on oscillatory neural networks (ONNs). We offer a first-of-its-kind and novel analog ONN computing architecture to seamlessly interface with sensors and process their analog data without any analog-to-digital conversion. ONNs are a biologically inspired neuromorphic computing architecture, where neuron oscillatory behavior will be developed by innovative phase change VO2 material coupled with synapses developed by bilayer Mo/HfO2 RRAM devices. PHASTRAC will address the most critical issues, namely 1) novel devices for implementing ONN architecture, 2) novel ONN architecture to allow analog sensor data processing, and 3) processing the data efficiently to take appropriate action. This “sensing-to-action” computing approach based on ONN technology will allow energy efficiency improvements of 100x-1000x and establish a novel analog computing paradigm for improved human-machine interactions.
The main objectives of PHASTRAC are:
(1) fabrication of low-power devices based on phase-change materials and bilayer metal-oxide RRAM devices that can serve as the building blocks for oscillatory neural networks (ONNs)
(2) development of devices, circuits to architecture-level models to allow exploration of the oscillatory neural network computing paradigm to allow the development of dedicated algorithms for ONN training
(3) development of a sensing-to-action computing paradigm with ONN that can serve as an energy efficient computing platform with potential interests in edge computing
(4) development of demonstrators and applications to show the advantage of ONN computing for AI tasks
In this first stage of the project, research efforts are dedicated to the starting of research activities for each task on the technical work packages.
The ongoing work in WP1 has been on the development of planar and crossbar VO2 devices as phase change material devices for oscillator implementation and bilayer metal oxide HfO2 RRAM devices as memristive devices for implementing synaptic coupling.
In WP2, ongoing work is focused on the development of models for understanding the device behavior and optimizing the devices in close exchange with experimentalists in WP1. In addition, there is ongoing research on developing circuit-to-architecture-level models for exploring the ONN design, performance, energy efficiency, and also its functionality in performing various tasks such as pattern recognition.
In WP3, the ongoing work is focused on algorithm development for training ONNs and enabling multi-modal learning capabilities. This work will bring together the models and insights from WP2 that can provide the specification and target for the demonstrator development in WP4.
In WP4, the ongoing work is on assessing various possible AI tasks that can be targeted via ONN.
The ongoing work in WP1 has been on the development of planar and crossbar VO2 devices as phase change material devices for oscillator implementation and bilayer metal oxide HfO2 RRAM devices as memristive devices for implementing synaptic coupling.
In WP2, ongoing work is focused on the development of models for understanding the device behavior and optimizing the devices in close exchange with experimentalists in WP1. In addition, there is ongoing research on developing circuit-to-architecture-level models for exploring the ONN design, performance, energy efficiency, and also its functionality in performing various tasks such as pattern recognition.
In WP3, the ongoing work is focused on algorithm development for training ONNs and enabling multi-modal learning capabilities. This work will bring together the models and insights from WP2 that can provide the specification and target for the demonstrator development in WP4.
In WP4, the ongoing work is on assessing various possible AI tasks that can be targeted via ONN.
PHASTRAC will bring innovation on devices, architecture, process integration and ONN applications. Here we describe each of them:
Advancement beyond State-of-the-Art on memristor devices
The VO2 material properties will be improved by engineering the ALD deposition process and implementing a flash-anneal technique to achieve better uniformity of the layers and control of composition and strain. In this project, the deposition method of bilayer metal oxide HfO2 devices will be realized directly on the clean interface of the first metallic layer, which is different from the reported processes up to now. Metals of choice in the first layer are gold, molybdenum or tungsten.
Advancement beyond State-of-the-Art on ONN Architecture
ONN state of the art will be advanced in two different directions: hardware implementations and learning. First, a fully digital implementation will be developed. Secondly, a hardware implementation using co-integration of novel devices with VO2 devices as neurons and bilayer MO HFO2 memristors as coupling elements is a significant step ahead with respect to state-of-the-art on energy efficient neuromorphic computing.
Advancement beyond the State-of-the-Art on ONN Applications
ONN will be able to target most neural network applications. We mainly focus on the applications geared from our partner AIM in Robotics based on single or multi-layer ONN by implementing multi-modal learning and capability of sense-to-compute architecture.
Advancement beyond State-of-the-Art on memristor devices
The VO2 material properties will be improved by engineering the ALD deposition process and implementing a flash-anneal technique to achieve better uniformity of the layers and control of composition and strain. In this project, the deposition method of bilayer metal oxide HfO2 devices will be realized directly on the clean interface of the first metallic layer, which is different from the reported processes up to now. Metals of choice in the first layer are gold, molybdenum or tungsten.
Advancement beyond State-of-the-Art on ONN Architecture
ONN state of the art will be advanced in two different directions: hardware implementations and learning. First, a fully digital implementation will be developed. Secondly, a hardware implementation using co-integration of novel devices with VO2 devices as neurons and bilayer MO HFO2 memristors as coupling elements is a significant step ahead with respect to state-of-the-art on energy efficient neuromorphic computing.
Advancement beyond the State-of-the-Art on ONN Applications
ONN will be able to target most neural network applications. We mainly focus on the applications geared from our partner AIM in Robotics based on single or multi-layer ONN by implementing multi-modal learning and capability of sense-to-compute architecture.