Variation-Adaptive Design in FPGAs

Lead Research Organisation: Imperial College London
Department Name: Electrical and Electronic Engineering

Abstract

This proposal is concerned with design techniques to compensate for parametric variations in deep-submicron configurable digital circuits. The growth of the microelectronics industry over the last four decades has been based on reducing the size of transistors to make integrated circuits smaller, faster and less expensive. The benefits obtained by transistor scaling are now diminishing as it becomes increasingly difficult to control the spread of parameters within a single chip.The research in this proposal will investigate new variation-aware design techniques in deep-submicron programmable logic devices. This is well-aligned with the recently developed Microelectronics Grand Challenges'' roadmap, particularly GC3---Moore for Less''. The premise for the research is that a single design can be implemented in more than one way in a given programmable logic device. In the presence of variations, distinct implementations will exhibit different performances, although they are functionally identical.The intention is to discover and develop strategies which exploit this feature in existing programmable logic devices. Furthermore, architectural modifications to such devices will be proposed to increase the obtainable benefit. These goals will be achieved from a thorough analysis of the nature of parametric variability from empirical studies, followed by the rigorous use of probabilistic modelling and optimisation theory.
 
Description New method to measure delay both online and offline. This results are the foundation of the EPSRC Programme Grant PRiME (EP/K034448/1).
Exploitation Route In addition to PRiME project, our method has also been used by groups around the world including U. Penn (DeHon), U. Sydney (Leong)
Sectors Digital/Communication/Information Technologies (including Software),Electronics
URL http://www.prime-project.org/2014/10/09/dvfs-demo/
 
Description Our findings in this project has sparked a new initiatives in researching into process variation in FPGAs and how this could be mitigated. Other groups that followed our work includes U. of York, Southampton, Newcastle, U. of Penn (USA), and U of Sydney and UNSW (Australia).
First Year Of Impact 2012
Sector Digital/Communication/Information Technologies (including Software),Electronics
Impact Types Economic
 
Description Altera 
Organisation Altera
Country United States of America 
Sector Private 
PI Contribution The research from the project has been followed up by Altera in investigating the impact of variation in FPGA to their future family. We have shared our results with their staff, in particular with Altera Fellow, Dr David Lewis.
Collaborator Contribution Altera provided us with hardware boards, FPGA chips and also they shared with us their confidential internal results on their measure variation data.
Impact Outcome includes a granted patent: P.Y.K. Cheung, N. P. Sedcole, and J. S. J. Wong. Method of measuring delay in an integrated circuit, Oct. 8 2013. US Patent 8,552,740.
Start Year 2010
 
Description Maxeler 
Organisation Maxeler Technologies
Country United Kingdom of Great Britain & Northern Ireland (UK) 
Sector Private 
PI Contribution We have assigned three of our patents to Maxeler: US6369610, US7543283, US12/747650
Collaborator Contribution They have provide us with hardware and software donations to support our projects.
Impact Impact as granted patents: US6369610, US7543283, US12/747650
Start Year 2010
 
Title Method of measuring delay in an integrated circuit 
Description A method of measuring signal delay in a integrated circuit comprising applying a common clock signal at a circuit input and output, applying a test signal at the circuit input, detecting a corresponding output signal at the circuit output and detecting whether the test signal and output signal occur in a common part of the clock signal. 
IP Reference US8552740 B2 
Protection Patent granted
Year Protection Granted 2013
Licensed Yes
Impact New methodology for measurement used by many.