Robust Enforcement of Customizable Resource Constraints in Heterogeneous Embedded Systems We develop an approach which enables embedded systems development with respect to customizable constraints on system-wide resource usage (time, energy, etc.). We utilize formal methods to develop a unified approach which addresses issues arising due to the trade-offs between computation, actuation, and sensing in embedded systems. This project is funded by National Science Foundation (NSF).
Smarter Nanoelectronics with Atomically Precise Graphene Nanoribbons (GNRs) We explore new ways to improve graphene nanoribbon (GNR) material synthesis involving ribbon length and to enhance single-device-level performance by exploiting novel GNR heterostructure systems (i.e., seamless lateral metal-semiconductormetal GNR junctions). This project is funded by Semiconductor Research Corporation (SRC).
Modeling the Memory-Compute Gap in Large-scale Superconductive Systems We develop circuit, microarchitectural simulation, and analytical models for the large-scale integration of Josephson-CMOS hybrid memories. This model will consider the high frequency nature of SFQ logic and explore strategies for improving memory interactions while reducing power consumption. This project is funded by the Department of Energy (DoE).
Low latency motion tracking for Autonomous Discovery Drones
Autonomous Discovery & Intercept missions can provide valuable intel against an adversary. While prior work has been completed into individual parts of such a system, there is no open established framework for this task. As such there is a need for a former framework to be developed, that can serve as the basis for further research. This project builds a framework for an autonomous interception drone. Included in this framework are a physical and simulated test platform as well as the detection and control software for this scenario.
CloudRenderVR: Motion-prediction based Speculative Cloud rendering for VR platforms
Graph-based analytical computation models for autonomous software and heterogeneous hardware
Scheduling in heterogeneous architectures:
Balancing performance/energy trade-off in energy-limited systems: We work on running neural network (NN) inference on multiple accelerators of an SoC. Our goal is to enable an energy-performance trade-off with an by distributing layers in a NN between a performance- and a power-efficient accelerator. We first provide an empirical modeling methodology to characterize execution and inter-layer transition times. We then find an optimal layers-to-accelerator mapping by representing the trade-off as a linear programming optimization constraint. We evaluate our approach on the NVIDIA Xavier AGX SoC with commonly used NN models. We use the Z3 SMT solver to find schedules for different energy consumption targets, with up to 98% prediction accuracy.
Multiple workload scheduling in heterogeneous architectures: We investigate Multi-accelerator Execution (MAE) on diversely heterogeneous embedded systems, where sub-components of a given workload, such as neural network inference, can be assigned to different type of accelerators to achieve a desired latency or energy goal. We first analyze the energy and performance characteristics of execution of neural network layers on different type of accelerators. We then explore energy/performance trade-offs via layer-wise scheduling for neural network inference by considering different layer-to-accelerator mappings.
Computation-Aware Planning in Autonomous Systems
Direct NVMe Accesses for GPU