For more information, visit the Configurable Computing Laboratory website.
Ever since the introduction of Field-Programmable Gate Array (FPGA) technology in the mid 80’s, people have imagined computers that could evolve and adapt to changing conditions without requiring external assistance. That dream has come closer to reality as ECE researchers have demonstrated the first known instance of a system changing its own hardware in a non-trivial fashion while continuing to run.
A team in the Configurable Computing Laboratory (CCM) has demonstrated an embedded system that is able to interactively implement new circuits inside itself or remove or reconnect existing circuits. The system was developed by Neil Steiner, a Bradley Fellow and Ph.D. candidate, working with his advisor, Peter Athanas.
By folding low-level design tools into a demonstration system, the team has made it possible for computers to participate in their own design. “If our system finds defects within itself, it can simply avoid using the damaged resources,” Steiner said. “This capability is much more cost-effective than throwing away large, but imperfect chips.”
Internal modeling and design tools also allow the system to function at a higher level of abstraction than previous FPGA systems. Instead of requiring a configuration bitstream that supports a specific FPGA model, the proof-of-concept system can accept circuit netlists in the widely popular EDIF format. These circuits are then placed, routed, configured, and connected to other circuits inside the system, without interrupting its operation. The same EDIF netlists can be used for any device in the same FPGA family, which ensures that designs remain more portable.
The current research falls within a proposed roadmap for autonomous computing systems—systems that are able to function more independently and assume responsibility for their own resources and operation, according to Steiner. The next step is to include a synthesizer within the system, so that it can accept behavioral circuit descriptions in high-level languages like VHDL or Verilog. “The most ambitious levels on the autonomy roadmap describe systems that can adapt to changing conditions and even learn from experience, a prospect that is not as far-fetched as it sounds,” he said.
Steiner credits his Bradley Fellowship and support from Xilinx, the leading FPGA manufacturer, in enabling his research to push the envelope to show what can be accomplished with current technology. “FPGAs have taken on important and well-deserved roles in modern systems, but we have shown that the early dreams are still viable,” he said.