Research
- Memcomputing devices
- Colloidal nanoparticle formation, its layer structure fabrication, device applications
- Integration of nanoparticles in nanotemplates and device applications
- Designed nanostructures for nanodevices
- Thin film deposition technology (PVD, CVD, ALD)
- Quantum dot formation through Stranski-Krastanow growth
- SiGe epitaxial layer for strained-Si device application
(1) Artificial Synaptic Devices for Neuromorphic Systems
The current computing systems have been successfully developed with a von Neumann architecture that is comprised of separated processing units and memory units. However, as the task for processing becomes more complex and the amount of data becomes more massive, they suffer from the high power consumption and delayed data transfer called von Neumann bottleneck. In order to overcome this limitation, the neuromorphic systems as a novel computing paradigm have been actively investigated. The neuromorphic systems emulate the human brain in computing, learning, training, and memory operation with high energy efficiency and low power. The artificial synapses emulating biological synaptic behavior as essential elements in neuromorphic systems have been under active investigation. They are connected to neuron circuits with a high connectivity and allow for simultaneous parallel signal processing and consequently adaptive learning/memory functions with a modulated synaptic weight. Our laboratory investigates the artificial synapses to develop neuromorphic systems realizing the brain operations.
- Memristors
Among the various synaptic devices developed for use in artificial synapses to date, memristors with two-terminal metal/resistive-switching-layer/metal structures, which are analogous to presynaptic-neuron/synapse/postsynaptic-neuron in biological systems, have been mostly actively investigated. Analog resistive switching memristors are particularly desirable to emulate the analog tuning of synaptic weight in biological synapses. Besides analog behavior, linearity of the resistance change, representing the identical incremental tuning of synaptic weight with the repetition of input pulses, is requisite for fast learning with simple neuron circuit operation by determining the synaptic weight change using only the pulse repetition number. The symmetric tuning of synaptic weight for synaptic potentiation and depression is also preferred because it allows the neuron circuit to generate voltage pulses with the same amplitude but opposite polarities for potentiation and depression. In addition, analog resistance changes need to display long-term stability to guarantee stable signal processing, adaptive learning, and memory functions. All of these characteristics should be obtained under the conditions of low power and low energy consumption. Our laboratory explores various synapse materials and device structures to realize these functions as artificial synapses for neuromorphic systems.
- Synaptic transistors
As discussed above, the synaptic device emulating biological synapse is the crucial unit for the neuromorphic system that potentially advances the electronic computing system with better energy efficiency, parallel operation with a high speed, and so on. This study is to investigate the synaptic materials and transistors emulating the function of biological synapse. Although the memristor-based devices have demonstrated the synaptic behaviors, it has remained to be challenging to realize the memristor with analogue resistance change that really emulates biological synapse. The biological synapse has an analogue change of synaptic strength, determined by the change of amount of ions and neuroreceptors, as the learning operation. This change retains as either short- or long-term memory. The changing and retaining connectivity of synapse affects the upcoming signal processing, which constitutes the adaptive operation. On the other hand, the fabricated memristors with metal-oxide and chalcogenide layers mostly exhibit digital-type resistance switching and only few groups including my group reported the analogue resistance change in memristor. Nevertheless, the stability and uniformity remain to be issues in an analogue memristor. Another issue is that the two-terminal memristor cannot perform learning and signal processing at the same time. The signal processing associates the signal transport from pre- to post-neuron by input pulse voltage on preneuron. On the other hand, the learning process, for example by spike-timing-dependent-plasticity (STDP), associates the sequential pulse at preneuron followed by fed-back pulse to synapse by postneuron with time difference. Thus, the signal processing would be aborted during learning operation.
However, the synaptic transistor can modulate the synaptic strength (conductance of transistor) for learning operation by applying gate voltage without disturbing the signal processing by pre- (source) and post-neuron (drain). Since the learning is updated by gate voltage during signal processing by source-drain voltage, the transistor has more flexibility to perform the learning and signal processing. The synaptic transistors should also have analogue conductance change in response to external stimuli, e.g. voltage or current, corresponding to change of synaptic strength in biological synapse. In this study, we pursue the analogue conductance change in semiconductor channel of synaptic transistor through redistribution of ions such as oxygen ions, charged vacancies, and protons, through ion exchange process with ionic conductor and storage layer. The ion exchange process is controlled by gate of transistor, and induces the biological synaptic function such as potentiation, depression, excitatory-postsynaptic-current, spike-timing-dependent-plasticity, etc. This study covers the concept of new synaptic transistors and synaptic materials including semiconductor channel having analogue conductance change determined by distributed ions, ionic conductors with controllable ionic transport by applied voltage with low power and high speed.
(2) Nonvolatile Memory Devices
Nonvolatile memories have played important roles in various electronic systems and recently attract further attention for emerging data-centric applications such as artificial intelligence that has to deal with vast amounts of data with high energy efficiency. Among various nonvolatile memories, the floating-gate memory has been most widely used as a unit cell of high-density nonvolatile flash memory. It has a simple and highly scalable metal-oxide-semiconductor field-effect transistor (MOSFET) structure with a floating-gate inside the gate oxide and operates with electrical charging of floating-gate (or charge-storage node), which alters the electrostatic potential in the channel layer and then shifts the threshold voltage. However, as the device scales down further, it suffers from undesirable and inevitable discharge over time as well as the statistical nature of electrical charging that causes the drift and distribution of threshold voltage and eventually leads to memory loss. In addition, the electrostatic crosstalk by stored charges between adjacent cells in the high-density memory architecture induces the threshold voltage drift, so limits further integration of memory cells. Therefore, it is essential to develop novel nonvolatile memory with the same MOSFET-based structure as utilizing the advantages of high-density architecture. Our laboratory explores novel nonvolatile memory devices by modulating gate stack properties to store the information such as memcapacitive response, channel conductance modulation, and so on, induced by voltage-driven modification of gate insulator and the semiconductor channel layer.
