Neuromorphic Device Integration & Characterization
Selector-Only Memory (SOM) represents a promising advancement in memory technology, offering solutions to some of the pressing issues faced by traditional memory devices. By leveraging the unique properties of OTS selectors, SOM can achieve high density, low power consumption, and fast performance, paving the way for the next generation of memory solutions. As research and development continue, overcoming the existing challenges will be key to unlocking the full potential of SOM technology, particularly in its critical role within the CXL platform and high-density memory applications.
Oxide-based two-terminal memristors serve a dual purpose by not only storing data within their analog conductance states but also processing data at the same physical locations. This unique capability enables highly efficient computing, both in-memory and in parallel. Our focus revolves around the design, integration, and thorough characterization of cutting-edge nanoelectronic devices tailored for the field of neuromorphic computing.
Phase-change memory (PCM) is a cutting-edge non-volatile memory technology based on the rapid and reversible phase transition between the amorphous and crystalline states of specific phase-change materials. This distinctive property of PCM, which allows for precise control of conductance levels, positions it as an ideal candidate for implementing synaptic functions in neuromorphic computing and memory applications.
The two-terminal Ovonic Threshold Switch (OTS) device has received extensive research attention, particularly for its application as a selector. This is primarily because it offers high current density for effectively programming and erasing memory elements when in the ON state. Moreover, it exhibits low leakage current when in the OFF state, making it ideal for preventing parasitic sneak currents in unselected cells within cross-bar arrays. One notable application of this technology is the implementation of Phase-Change Memory (PCM) with OTS elements, often referred to as PCMS. This approach has been put into practice by Intel, particularly for high-density memory applications, further underscoring the significance of the OTS device in advanced memory technology.
Device simulation
We have developed a comprehensive model to explain the resistive switching (RS) behavior in memristors and have leveraged it to propose enhanced device structures. Two distinct filament growth modes have been identified, each giving rise to varying slopes of conductance modulation and dynamic ranges. Understanding these different modes and their associated observations will play a crucial role in ongoing efforts to optimize devices, paving the way for their application in memory and low-power neuromorphic computing applications.
Neuromorphic computing system (Deep learning accelerator)
Our research is centered on the development of a fully integrated, reprogrammable memristor chip. This chip includes a passive memristor crossbar array seamlessly integrated with all essential interface circuitry, digital buses, and a processor, creating a comprehensive neuromorphic hardware system-on-chip.
The memristor array's exceptional high density and non-volatile characteristics enable the on-chip storage of entire Deep Neural Network (DNN) models. This effectively eliminates the inefficiencies associated with off-chip memory access, leading to the promise of significantly improved energy efficiency in DNN operations.