VLSI Systems

IBM Research has a long history of innovation in the development of high-speed VLSI devices for modern computing systems. Today, while we continue to find ways to increase raw processing speed, we are also creating new systems optimized for power consumption and data movement.

Advances in silicon technology continue to drive rapid improvements in the performance and functionality of VLSI systems. Microprocessor transistor counts and clock frequencies, for instance, regularly double every 24 months. New designs and architectural enhancements allow overall microprocessor performance to increase even more rapidly, doubling every 16 months. High-end microprocessor designs currently contain well over 100 million transistors and operate at speeds greater than 1 GHz. These designs are really VLSI systems with 10 or more execution units, several levels of cache memory, and a rich set of I/O interfaces. Such microprocessors are an important element in IBM's VLSI systems portfolio; our zSeries™ and PowerPC® microprocessors are key to their respective product lines. We continue to make major contributions to IBM's products and strategy in this area through research projects and development partnerships addressing advanced microarchitectures, circuit techniques, design tools and methodologies, packaging, cooling, and testing and analysis. In the VLSI circuits area, we are developing new circuit families, both synchronous and asynchronous, with these attributes.

The same silicon-technology trends that drive advanced microprocessors also affect a rich array of other VLSI systems. For example, IBM has become the world's leading application-specific integrated circuits(ASIC) provider by leveraging our system-on-a-chip (SOC) design methodology. This methodology and its companion IP core library are being used and extended by several research projects. The Earl ultra-low-power project is extending our SOC capabilities into the battery-powered space through new voltage- and frequency-scaling circuit techniques. We are developing a number of approaches to reducing power consumption: high-performance, low-voltage circuits that control leakage current; dynamic voltage- and frequency-scaling techniques; voltage-regulation and level-shift methods; and optimization of the hardware/software interface with deep-sleep modes that have near-zero power dissipation. These innovations will have application in non-battery environments as well, such as super-dense compute servers for web hosting.

We are also looking at applications of VLSI for high-density computing systems and high-bandwidth data communications. The Blue Gene project will integrate numerous custom VLSI chips, each with multiple compute modules having processor, memory, and communications, to construct a system with one million processors. Our super-dense servers for compute-farm applications will employ low-power, embedded processors built with SOC methodologies to maximize MIPS per volume under constraints on system power dissipation and temperature control.

Rapid increases in communication bandwidth demand increasing performance from special-purpose compute engines called network processors, which are systems optimized for functions like packet classification and packet forwarding. These systems are particularly challenging due to their demands on bandwidth, cost, and programmability. Beyond the speed requirements for packet processing, we are also researching critical security technology for VLSI network processors.

Please contact Paridhi Verma to obtain copies of the Computer Science Brochure