|
 Imperial College has one of the largest and strongest groups in the world on custom computing, field-programmable logic, system-on-chip technologies, hardware/software codesign, and reconfigurable systems. Led by six members of academic staff, they have been responsible for many innovations, including theory and practice of hardware compilation, models for run-time reconfigurable designs, bit-width optimization techniques, novel design languages and transformation methods, and domain-specific hardware platforms. Their research has been supported by many institutions, including UK Research Councils, European Union, Altera, Celoxica, Cisco, Hewlett Packard, Morgan Stanley, Sharp, Sony, and Xilinx. Computer Systems Group Websites: Research The main research focus is on optimisation of performance, productivity and power consumption in designing hardware/software systems. The work includes theory and practice of custom instruction sets, hardware accelerators (e.g. based on FPGAs), static and dynamic architecture and code optimisation, as well as dynamic code generation. The section has strong links with industry and collaborative projects with the EE Department. - Hardware + Software Systems: Theory + Practice
- P³ Optimisations: Performance, Power, Productivity
Methods - Processor custom instructions
- Generation of hardware accelerators
- Static and dynamic code optimisation
- Run-time code generation
Research Results Performance and Power Consumption: - Elliptic Curve Encryption: 1150x speed - parallel+hardware
- Pipelining: 90% power reduction
- Distributed Java (RMI): 4.6x speed - reduce communication
- FIP: customisable instruction processor (patented)
Productivity: methods and tools: - ASC: A Stream Compiler, rapid h/w design exploration
- Haydn-C, Quartz: parallelism in hardware, multi-level and user-directed
- TGL: Task Graph Library for dynamic s/w optimisation
|
