The confluence of Big Data, IoT and Real-Time Analytics calls for rethinking of the hardware computing paradigm, either by the bottom up or top down approach. Memristive neuromorphic systems inspired by brain functions and implemented through new materials properties, bionic memristive devices (e.g., artificial synapses and neurons) and neural network circuits, are emerging with the promise of transforming the semiconductor information processing technology.
Since the experimental discovery of memristors twelve years ago, memristive neuromorphic hardware has continued to make big strides. On the level of the material, a huge number of materials of various categories have shown memristive properties and this number continues to increase rapidly. In such a growing filed, there is a crucial need to establish materials selection rules and evaluate the suitability of the corresponding devices for neuromorphic systems. Device yield testing, performance distribution analysis and circuit reliability simulations have to be performed and standardized to turn these materials research endeavors into real impact. On the device level, researchers have showcased devices with diverse physical mechanisms, primary neuromorphic functions, and intriguing performances. Highly reproducible devices with desired characteristics, and versatile devices integrating multiple neural dynamics and computational capabilities remain to be demonstrated to further boost the system performance and functionality. On the circuit level, many progresses have already been achieved from a few tens of devices-scale to mega-scale, from single-array to multi-array functional circuits benchmarked by executing typical machine learning tasks and basic neural algorithms. Even larger and more compact circuits are necessary for real-world applications. Meanwhile, it is also tempting to consider the memristive circuits not only as the accelerators of machine learning algorithms but also as platforms for unprecedented physical computing which harvest many attractive physical properties of the hardware systems, such as device nonideality, stochasticity and emergent phenomena, many of which can find their neuronal analogues. Communication and co-design across these different levels are also essential to fully unleash the potential of the memristive neuromorphic systems.
Topics of interest include, but are not limited to:
1. Rational design or selection of memristive materials with materials-device-circuit-architecture-algorithm codesign considerations or statistical analysis of the experimental data.
2. Proof-of-concept neuromorphic memristive devices with enhanced bio-fidelity and neuronal computation capabilities.
3. High-yield memristive device technologies, high-performance and highly reliable memristive devices for integrated circuits.
4. Proof-of-concept memristive circuit functionalities that take advantage of the natural properties of the hardware.
The confluence of Big Data, IoT and Real-Time Analytics calls for rethinking of the hardware computing paradigm, either by the bottom up or top down approach. Memristive neuromorphic systems inspired by brain functions and implemented through new materials properties, bionic memristive devices (e.g., artificial synapses and neurons) and neural network circuits, are emerging with the promise of transforming the semiconductor information processing technology.
Since the experimental discovery of memristors twelve years ago, memristive neuromorphic hardware has continued to make big strides. On the level of the material, a huge number of materials of various categories have shown memristive properties and this number continues to increase rapidly. In such a growing filed, there is a crucial need to establish materials selection rules and evaluate the suitability of the corresponding devices for neuromorphic systems. Device yield testing, performance distribution analysis and circuit reliability simulations have to be performed and standardized to turn these materials research endeavors into real impact. On the device level, researchers have showcased devices with diverse physical mechanisms, primary neuromorphic functions, and intriguing performances. Highly reproducible devices with desired characteristics, and versatile devices integrating multiple neural dynamics and computational capabilities remain to be demonstrated to further boost the system performance and functionality. On the circuit level, many progresses have already been achieved from a few tens of devices-scale to mega-scale, from single-array to multi-array functional circuits benchmarked by executing typical machine learning tasks and basic neural algorithms. Even larger and more compact circuits are necessary for real-world applications. Meanwhile, it is also tempting to consider the memristive circuits not only as the accelerators of machine learning algorithms but also as platforms for unprecedented physical computing which harvest many attractive physical properties of the hardware systems, such as device nonideality, stochasticity and emergent phenomena, many of which can find their neuronal analogues. Communication and co-design across these different levels are also essential to fully unleash the potential of the memristive neuromorphic systems.
Topics of interest include, but are not limited to:
1. Rational design or selection of memristive materials with materials-device-circuit-architecture-algorithm codesign considerations or statistical analysis of the experimental data.
2. Proof-of-concept neuromorphic memristive devices with enhanced bio-fidelity and neuronal computation capabilities.
3. High-yield memristive device technologies, high-performance and highly reliable memristive devices for integrated circuits.
4. Proof-of-concept memristive circuit functionalities that take advantage of the natural properties of the hardware.
