Home      People     Projects     Publications    VSL 

     
 
NASICs
(Nanoscale Application Specific ICs and Architectures)
  • we are designing circuits and architectures based on carbon nanotubes (CNT) and silicon nanowires (SiNW)
  • we have published several papers recently (see our publications for details)
  • collaborating with several device physics groups (e.g., Mark Tuominen's group)

Cool-* (back to top)

Cool-* (read CoolStar) is our flagship research; it is a, statically speculative, low-power microprocessor architecture. Our goal is to reduce power and energy consumption by a suite of combined architecture-compiler techniques. Our main philosophy is leveraging static information speculatively and providing support for static execution modes in addition to other circuit and architecture level techniques. We estimate that our approach can reduce power (energy) by 40-70% of a state-of-the-art low-power microprocessor such as the StrongARM. We have developed techniques in different areas of the processor, the Cool-* suite, e.g., Cool-Cache (see Micro-34, Dec 2001), Cool-Fetch (ACM Computer Architecture Letters, 2002, May), Cool-Mem (to appear in ASPLOS 2002), Cool-Run (in progress), Cool-Ways(submitted), Cool-Watts (in progress), MiniMax Cache (in HPCA 2002)). Check the list of publications for already published works and check back later because we have many interesting approaches cooking in the pipeline .....

FlexCache (back to top)

A fully compiler-enabled memory system. It integrates software cache partitions, compiler managed partitions, and conventional cache partitions into a uniform memory system framework. We plan to target FlexCache to:
  • synthesize caches for reconfigurable architectures,
  • provide predictable memory systems in real-time embedded systems
  • improve caching efficiency  in microprocessors
  • reduce power consumption in memory systems.
     
Synchronization Coherence Protocols  (back to top)
  • integrating cache coherence with fine grained synchronization in SMP/NUMA/single-chip-multiprocessors
  • we have built a multiprocessor simulator that supports synchronization coherence up to 512 nodes
  • collaboration with Vlad Vlassov (KTH Stockholm) and Richard Weiss (Hampshire College)

Euphoria    (back to top)
EuphoriaLite focuses on scalable software infrastructure for global dynamic content management applications on cluster of workstations and multiprocessors, as well as in distributed embedded systems.  Euphoria  implements it's scalable/adaptive resource management based on:
  • non-intrusive introspection byte code level analysis and transformations (Java and .NET CLR),
  • individualized content management,
  • object cache store management,
  • analytical forecasting models.
  • web caching
  • wireless client integration HTML/XML-WAP
  • real-time JVM for resource constrained devices

From the outside it aims to be compatible with  application server and client software technologies such as defined in J2EE, J2ME, CLDC, WAP, KVM specifications and have support for both traditional web based and wireless clients.


Software virtualization techniques
  • Hot Pages
  • Compiler Managed Memory

Analytical and Compiler Generated Simulation of Billion Transistor Architectures (back to top)
  • Integrated framework for cost performance valuation of architectures with 1000s of processing units
  • SimpleFit - analytical modeling of Raw and FPGAs



 
© 2005 UMass Amherst. Site Policies