Neuron & Synapse

Artificial intelligence (AI) emulating biological neural networks has been investigated for decades, and it is noticeable that the practical applications of AI, which require the processing of large amounts of data and thus place significant demands on computing speed and power efficiency, are limited by computing hardware. The conventional von Neumann architecture suited to scientific computing is inefficient in handling AI tasks due to the bottleneck of data transfer.

Fortunately, the advent of memory resistive devices with innate dynamics similar to biological synapses or neurons provides a viable and simple way to build hardware-based biomimetic computing systems.

In our laboratory, researchers are developing many different synaptic and neuron devices in noble materials for various applications.

Integration

As CMOS approaches its physical limits, ENVM-based neuro-inspired computing chips offer a promising future.

Due to device development efforts, the on-chip capacities of synapses and neurons have increased significantly in recent years. However, the cell density has not been efficient for AI processing due to the lack of device performance.

Future device studies should focus on not only the capacities of the devices but also on implementing analog NVMs so that we can achieve improved and more biologically plausible.

with high-density 3D NVMs integration, future neural-inspired computing chip integration will be more efficient in terms of area and our goal is to reach much larger integration scales.

Noble Security Device

With the rapid development of electronic device technology that occupies a large part of our daily life, security is the most pressing need of all emerging technologies.

The unique probabilistic characteristic of memristors due to dynamics of ions such as resistance variability, random telegraph noise (RTN), and stochastic switching is suitable to use as a security device.

Our group is demonstrating the in-memory computing system for PUF (Physical Unclonable Function) which can provide a unique hardware identity, and TRNG (True Random Number Generator).