A5020136488- Advanced Memory and Neural Computing
- Semiconductor materials and devices
- Ferroelectric and Negative Capacitance Devices
- Energy Harvesting in Wireless Networks
- Parallel Computing and Optimization Techniques
- Neural Networks and Reservoir Computing
- Advancements in Semiconductor Devices and Circuit Design
- Phase-change materials and chalcogenides
- Transition Metal Oxide Nanomaterials
- MXene and MAX Phase Materials
- Advanced Neural Network Applications
- Machine Learning and Algorithms
- Domain Adaptation and Few-Shot Learning
- CCD and CMOS Imaging Sensors
- Thin-Film Transistor Technologies
- Interconnection Networks and Systems
- Low-power high-performance VLSI design
- 3D IC and TSV technologies
Pennsylvania State University
2016-2021
Park University
2017
Brain-inspired cognitive computing has so far followed two major approaches - one uses multi-layered artificial neural networks (ANNs) to perform pattern-recognition-related tasks, whereas the other spiking (SNNs) emulate biological neurons in an attempt be as efficient and fault-tolerant brain. While there been considerable progress former area due a combination of effective training algorithms acceleration platforms, latter is still its infancy lack both. SNNs have distinct advantage over...
Many recent works have shown substantial efficiency boosts from performing inference tasks on Internet of Things (IoT) nodes rather than merely transmitting raw sensor data. However, such tasks, e.g., convolutional neural networks (CNNs), are very compute intensive. They therefore challenging to complete at sensing-matched latencies in ultra-low-power and energy-harvesting IoT nodes. ReRAM crossbar-based accelerators (RCAs) an ideal candidate perform the dominant...
This paper presents the first monolithic 3D vertical cross-tier computing-in-memory (CIM) SRAM cell fabricated using low cost TSV-free FinFET-based <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">+</sup> -IC technology. The 9T CIM is able to compute NAND/AND, OR/NOR and XOR/XNOR operations within a single memory cycle. We stackable multi-fin single-grained Si FinFET thermal-budget CO <inf xmlns:xlink="http://www.w3.org/1999/xlink">2</inf>...
We present an extensive analysis of functional-oxide based selector devices for cross-point memories from the perspectives materials through arrays. describe design constraints required proper functionality a array and translate these to figures merit materials. The proposed merit, related resistivities functional oxide in metallic insulating states critical current densities insulator-metal transitions, determine whether or not is suitable be employed as memory technology. Our shows...
We perform a simulation-based analysis on the potential of emerging ferroelectric tunnel junctions (FTJs) as memory device for crossbar arrays. Though FTJs are promising due to their low power switching characteristics compared other technologies, greatest challenge is tradeoff between integration density and read performance. Our highlights need co-optimize thickness FTJ read/write voltages achieve proper functionality at large array sizes. shows that FTJ-based achieves 93% higher sense...
With data volume growing exponentially in today's era, modern computing systems are increasingly bottlenecked and consistently burdened by the costs of movement. Driven development emerging non-volatile memory (NVM) technologies increasing demand for high throughput big applications, considerable research effort has gone into embedding exploiting parallelism data-intensive workloads to address "memory wall" bottleneck. In this work, we propose a design which leverages run-time...
Compute-in-Memory (CiM) techniques focus on reducing data movement by integrating compute elements within or near the memory primitives. While there have been decades of research various aspects such logic and integration, confluence new technology changes emerging workloads makes us revisit this design space. This work focuses functionality that can be embedded to SRAMs using monolithic 3D integration. Properties transform costs embedding compared prior efforts. also explores how into...
There is an ongoing trend to increasingly offload inference tasks, such as CNNs, edge devices in many IoT scenarios. As energy harvesting attractive power source, recent ReRAM-based CNN accelerators have been designed for operation on harvested energy. When addressing the instability problems of energy, prior optimization techniques often assume that load fixed, overlooking close interactions among input power, computational load, and circuit efficiency, or adapt dynamic match just-in-time...
Recent advances in emerging technologies such as monolithic 3D Integration (M3D-IC) and non-volatile memory (eNVM) have enabled to embed logic operations memory. This alleviates the "memory wall" challenges stemming from time power expended on migrating data conventional Von Neumann computing paradigms. We propose a M3D SRAM dot-product engine for compute in-SRAM support used applications matrix multiplication artificial neural networks. In addition, we novel RRAM-based architecture...
Cross-point architecture, while being appealing in consideration of high integration density, suffers from leakage through sneak paths across the array. The current flowing half-accessed and some cases, unaccessed cells (and corresponding power) are important determinants array performance. Proper estimation these components is computationally challenging often demands rigorous simulation efforts. This paper presents a efficient compact model to assess cross-point employing threshold switch...
Conventional processors suffer from high access latency and power dissipation due to the demand for memory bandwidth data-intensive workloads, such as machine learning analytic. In-memory computing support various technologies has provided formidable improvement in performance energy alleviating repeated accesses data movement between CPU storage. While many processing in-memory (PIM) works have been proposed efficiently compute dot products using Kirchoff's law, solutions are unsuitable...
This work will provide an overview of recent advances in enabling SRAM-based compute fabrics leveraging monolithic 3D (M3D). It highlight that the fine grain connectivity enabled by M3D, enables to embed computations close memory cells significantly reducing data transfer costs. The application level benefits emerging workloads also be presented.