A5020900113- Parallel Computing and Optimization Techniques
- Energy Efficient Wireless Sensor Networks
- Embedded Systems Design Techniques
- Advanced Memory and Neural Computing
- Ferroelectric and Negative Capacitance Devices
- Context-Aware Activity Recognition Systems
- Mobile Ad Hoc Networks
- Modular Robots and Swarm Intelligence
- VLSI and Analog Circuit Testing
- Advanced Data Storage Technologies
- Indoor and Outdoor Localization Technologies
- Advanced Image Processing Techniques
- Particle Detector Development and Performance
- Advanced Vision and Imaging
- Wireless Networks and Protocols
- Opportunistic and Delay-Tolerant Networks
- Low-power high-performance VLSI design
- Interconnection Networks and Systems
- RNA modifications and cancer
- Botulinum Toxin and Related Neurological Disorders
- VLSI and FPGA Design Techniques
- Millimeter-Wave Propagation and Modeling
- Security in Wireless Sensor Networks
- Nerve injury and regeneration
- Calpain Protease Function and Regulation
University of California, Berkeley
2005-2024
University of Maryland, Baltimore
2024
Berkeley College
2019-2024
University of California, San Diego
2023
National Institutes of Health
2021
National Cancer Institute
2021
Center for Cancer Research
2021
New York Institute of Technology
2020
University of Cambridge
2018
University of Utah
2013
Wireless sensor networks promise fine-grain monitoring in a wide variety of environments. Many these environments (e.g., indoor or habitats) can be harsh for wireless communication. From networking perspective, the most basic aspect communication is packet delivery performance: spatio-temporal characteristics loss, and its environmental dependence. These factors will deeply impact performance data acquisition from networks.In this paper, we report on systematic medium-scale (up to sixty...
As new fabrication and integration technologies reduce the cost size of micro-sensors wireless interfaces, it becomes feasible to deploy densely distributed networks sensors actuators. These systems promise revolutionize biological, earth, environmental monitoring applications, providing data at granularities unrealizable by other means. In addition challenges miniaturization, system architectures network algorithms must be developed transform vast quantity raw sensor into a manageable...
Wireless sensor networks promise fine-grain monitoring in a wide variety of environments. Many these environments (e.g., indoor or habitats) can be harsh for wireless communication. From networking perspective, the most basic aspect communication is packet delivery performance: spatio-temporal characteristics loss, and its environmental dependence. These factors will deeply impact performance data acquisition from networks.In this paper, we report on systematic medium-scale (up to sixty...
Continued improvement in computing efficiency requires functional specialization of hardware designs. Agile design methodologies have been proposed to alleviate the increased costs custom silicon architectures, but their practice thus far has accompanied with challenges integration and validation complex systems-on-a-chip (SoCs). We present Chipyard framework, an integrated SoC design, simulation, implementation environment for specialized compute systems. includes configurable, composable,...
DNN accelerators are often developed and evaluated in isolation without considering the cross-stack, system-level effects real-world environments. This makes it difficult to appreciate impact of Systemon-Chip (SoC) resource contention, OS overheads, programming-stack inefficiencies on overall performance/energy-efficiency. To address this challenge, we present Gemmini, an open-source, full-stack accelerator generator. Gemmini generates a wide design-space efficient ASIC from flexible...
Wireless sensor networks involve very large numbers of small, low-power, wireless devices. Given their unattended nature, and potential applications in harsh environments, we need a monitoring infrastructure that indicates system failures resource depletion. We describe an architecture for network monitoring, then focus on one aspect this architecture: continuously computing aggregates (sum, average, count) properties (loss rates, energy-levels etc., packet counts). Our contributions are...
We propose a practical and scalable technique for point-to-point routing in wireless sensornets. This method, called Beacon Vector Routing (BVR), assigns coordinates to nodes based on the vector of hop count distances small set beacons, then defines distance metric these coordinates. BVR routes packets greedily, forwarding next that is closest (according this beacon metric) destination. evaluate approach through combination high-level simulation investigate scaling design tradeoffs,...
Recent technological advances and the continuing quest for greater efficiency have led to an explosion of link network protocols wireless sensor networks. These embody very different assumptions about stack composition and, as such, limited interoperability. It has been suggested [3] that, in principle, networks would benefit from a unifying abstraction (or "narrow waist" architectural terms), that this should be closer level than level. This paper takes vague principle turns it into...
As new fabrication and integration technologies reduce the cost size of micro-sensors wireless interfaces, it becomes feasible to deploy densely distributed networks sensors actuators. These systems promise revolutionize biological, earth, environmental monitoring applications, providing data at granularities unrealizable by other means. In addition challenges miniaturization, system architectures network algorithms must be developed transform vast quantity raw sensor into a manageable...
This paper presents a novel runtime power modeling methodology which automatically identifies key signals for dissipation of any RTL design. The toggle-pattern matrix is constructed with the VCD dumps from training set, where each signal represented as high-dimensional point. By clustering showing similar switching activities, small number are selected, and then design-specific but workload-independent activity-based model using regression against cycle-accurate traces obtained...
Serialization frameworks are a fundamental component of scale-out systems, but introduce significant compute overheads. However, they amenable to acceleration with specialized hardware. To understand the trade-offs involved in architecting such an accelerator, we present first in-depth study serialization framework usage at scale by profiling Protocol Buffers ("protobuf") across Google's datacenter fleet. We use this data build HyperProtoBench, open-source benchmark representative key...
General-purpose lossless data compression and decompression ("(de)compression") are used widely in hyperscale systems key "datacenter taxes". However, designing optimal hardware processing units ("CDPUs") is challenging due to the variety of algorithms deployed, input characteristics, evolving costs CPU cycles, network bandwidth, memory/storage capacities.
In response to growing application diversity, System-on-Chip (SoC) architectures have become increasingly heterogeneous, with diverse cores and accelerators, as well non-uniform memory systems. However, existing open-source design frameworks for SoCs NoCs (Network-on-Chips) been unable facilitate exploration of heterogeneous SoC irregular NoCs. We present Constellation, a new NoC RTL generator framework designed from the ground up support integration in evaluation highly architectures....
With the widespread adoption of deep neural networks (DNNs) across applications, there is a growing demand for DNN deployment solutions that can seamlessly support multi-tenant execution. This involves simultaneously running multiple workloads on heterogeneous architectures with domain-specific accelerators. However, existing accelerator interfaces directly bind accelerator's physical resources to user threads, without an efficient mechanism adaptively re-partition available resources. leads...
Wireless sensor networks have been attracting increasing research interest given the recent advances in miniaturization and low-cost, low-power design. Consisting of a large collection small wireless, low-power, unattended sensors and/or actuators, wireless network technology poses its unique design challenges. Given their nature complexity, it is critical that users be continuously updated indications health, i.e., explicit knowledge overall state after deployment. We call such health...
This work presents a 16mm <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> heterogeneous RISC-V system-on-a-chip (SoC) composed of high-performance out-of-order core, energy-efficient in-order data-parallel vector accelerator, and systolic array deep neural network (DNN) accelerator in low-power Intel 22FFL for general-purpose compute, DNN, workloads. The SoC is fully open-source components, including second-generation Berkeley...
Timing and pattern of mitochondrial potential (m) depolarization during no-flow ischaemia-reperfusion (I-R) remain controversial, at least in part due to difficulties interpreting the changes fluorescence m-sensitive dyes such as TMRM. The objective this study was develop a new approach for confocal TMRM signals I-R based on spatial periodicity packaging ventricular cardiomyocytes. (FTMRM) recorded from Langendorff-perfused rabbit hearts immobilized with blebbistatin using either microscope...