The Field Programmable Logic Perspective
Senior Director, Xilinx Research Labs
The economics of the semiconductor industry now dictate that only a small number of elite companies can justify
implementing their products and designs directly on the most advanced silicon processes. One consequence is that the business and technical assumptions that originally gave rise to the invention of Xilinx Configurable Logic Arrays (now called FPGAs) are increasingly relevant for many other semiconductor companies. For example, it is now the norm to rely on a fabless or “fab-lite” business model and to build programmable architectures whose design costs can be amortized across as many potential application markets as possible. The widespread re-emergence of parallel programmable architectures, such as multi-core processors or GPUs, is the industry’s response to the “power wall” that has halted the progress of ever-faster microprocessors. These programmable architectures seek performance by trading parallelism for execution speed. Ironically, after years of investment in design abstractions, programming models and languages, the practice of software design is starting to look more like hardware design. Both are converging
on a common set of challenges typified by issues such as spatial partitioning and power management.
The future of digital system design is increasingly parallel and programmable.
Additionally, heterogeneous compute resources and architectural adaptivity are important vectors in the design space. Compounding these trends, power management is consistently the most critical design constraint that intersects all other concerns. Collectively, the construction of correct designs in a cost-effective manner is the industry’s primary challenge.
In such a landscape programmable logic is set to play an increasingly important role.
In this context, it is instructive to review past and current research efforts.
It is challenging to assess the prospects for breakthroughs that will relieve the bottlenecks of design and
verification productivity. Finally, we speculate on some novel approaches which offer potential but have not yet been thoroughly explored.
Patrick Lysaght is a Senior Director in Xilinx Research Labs, San Jose, Ca.
He leads a group whose research interests include power estimation, system-level modeling, reconfigurable computing (especially dynamically reconfigurable systems) and emerging design technologies for FPGAs. He also directs the worldwide operation of the Xilinx University Program (XUP).
Before joining Xilinx, he held positions as a senior lecturer at the University of Strathclyde (Glasgow) and at the Institute for System Level Integration (Livingston, Scotland). He started his career in research and development with Hewlett Packard (Edinburgh) before going on to hold a number of technical and marketing positions.
Patrick has co-authored more than fifty technical papers, co-edited two books on programmable logic and holds six US patents. He is actively involved in the organization of a number of international conferences and is chairman of the steering committee for FPL, the world’s largest conference dedicated to field programmable logic. Patrick holds a BSc (Electronic Systems) from the University of Limerick, Ireland and a MSc degree (Digital Techniques) from Heriot-Watt University in Edinburgh, Scotland.
He is a Visiting Professor at Beijing University of Technology, China and the University of Strathclyde, Glasgow, UK.