Hundreds of engineers gathered at the DARPA Electronics Resurgence Initiative Summit in San Francisco yesterday to hear that dozens of them were getting millions (one group was awarded more than US $60 million) to tackle some big questions. If researchers can answer them in the affirmative, they will keep electronics going long after Moore’s Law is a thing of the past. And in the process they’ll likely change the nature of the industry as well as the jobs of engineers.
The Electronics Resurgence Initiative has three main thrusts: design, architecture, and materials and integration. (For a deep dive into the programs and the reasoning behind them, see last week’s interview with the initiative’s leader, Bill Chappell.)
In the design field, DARPA’s Microsystems Technology Office deputy director, Jay Lewis, announced the funding for projects under two efforts, IDEA and POSH. The two programs seek to usher in an era of the 24-hour design cycle for defense-related hardware systems, shorten the time it takes to upgrade hardware, and enable the proliferation of custom systems-on-chip.
IDEA aims to create a “no human in the loop” layout generator that would enable users with even limited electronic design expertise to complete the physical design of electronic hardware within 24 hours. The biggest award in that sector ($24.1 million) went to a group led by David White at the electronic design automation firm Cadence, but including several partners. “This program will set the stage for enhancing the entire span of our analog, digital, verification, package and PCB EDA technologies, providing our customers with the most advanced system design enablement solutions,” said Anirudh Devgan, president of Cadence, in a statement. In an address at the Summit, Cadence senior vice president Tom Beckley pointed out that the company’s premier EDA platform, Virtuoso, already includes machine-learning technology, and they have about 30 engineers working on machine learning. The IDEA project will help them add to that number.
Principal | Funds (millions of US$) | Institutions | Description |
David White | 24.1 | Cadence, Nvidia, University of California Berkeley, Univeristy of Texas Austin, Carnegie Mellon University | A no-human-in-the-loop layout generator for analog-, package-, and board design |
Andrew Kahng | 11.3 | University of California, San Diego; Qualcomm, ARM, University of Minnesota, Brown University, University of Michigan, Univerisity of Illinois Urbana-Champaign; University of Texas, Dallas | Open source layout generator for digital circuits, packages, and boards through the use of extreme partitioning, cloud computing, and machine learning |
Jonathan Bachrach | 8.7 | Northrop Grumman, JITX | An open source system generator, open parts data base, and circuit optimizer to enable automatic design of boards and packages baseed on user design intent |
David Wentloff | 6.4 | University of Michigan, ARM, Univeristy of Virgina. | Automated SoC synthesis tool for both analog and digital components from a user-intent-based description |
Sachin Sapatnekar | 5.3 | University of Minnesota, Intel, Texas A&M | Open source layout generator for analog circuits throgh the use of template-drive designs and machine learning |
David Wentzlaff | 2.8 | Princeton University, University of Washington | Develop open-source high-imapct digital cicuit, analgo circuit, board, and package validation test cases and evaluate the IDEA design flow |
Martin Wong | 1.7 | University of Illinois | Open source massively parallel EDA infrastructure and stati timing analysis engine |
Nan Sun | 1.7 | University of Texas | Machine learning-based software for analog layout constraint generation, placement, and routing that leverages human-expert constraints derived from existing analog layouts |
Dan Jiao | 1.3 | Purdue University | First principle physics-based extraction flow for PCBs, packages, and System-on-Chips |
Rajit Manohar | 1.2 | Yale University, UC Berkeley, UT Austin, Sandia National Labs | Develop an open-source EDA flow for asynchronous circuits |
Pierre-Emmanuel Gaillardon | 1 | Univeristy of Utah | Open source logic sysnthesis leveraging logic optimization engines and machine-learning-based logic synthesis technqiues based on input circuit topology |
The othe design effort, POSH (for Posh Open Source Hardware) aims to bring the culture and capability of open source to hardware design. “To help democratize access to custom, high-performance SoCs, the POSH program seeks to develop a sustainable ecosystem of open-source IP and accompanying validation tools,” said Andreas Olofsson, the program manager leading IDEA and POSH, in a statement. “Through POSH, we hope to eliminate the need to start from scratch with every new design, creating a verified foundation to build from while providing deeper assurance to users based on the open source inspection process.”
Principal | Funds (millions of US$) | Institutions | Description |
Eric Keiter | 6.9 | Sandia National Labs, Yale University, University of California Berkeley | Massively parallel distributed open source simulation tools |
Alex Rabinovitch | 6.1 | Synopsys, Lockheed Martin, Analog Devices, Analog Circuit Works | An emulation platform for high-performance analog mixed signal functional verification with transistor-level accuracy |
Tony Levi | 6 | University of Southern California, GlobalFoundries | High-performance open-source analog circuits |
Clark Barrett | 5.9 | Stanford University, Princeton University | Open source model checkers and instruction-level abstractions for arbitraray hardware and mixed-signal models for formal verification of analog IP blocks |
Michael Taylor | 2.7 | Univeristy of Washington, Boston University | High-performance open source 64-bit RISC-V multicore processor |
Richard C.J. Shi | 2.5 | University of Washington | PDK-independent topology description language for the deliveriy of open source mixed signal IP modules portable from 180-nm to 14-nm technology nodes |
David Wentzlaff | 1.8 | Princeton University, Cornell University | Open-source FPGA fabric and scalable coherence memory system for creation of an open source hardware ecosystem |
Edgar Inglesias | 1.2 | Xilinx, GreenSoCs, FEIMTECH AB | An open source appraoch for fast mixed-accuracy cosimulation that allows hardware and software to be co-developed, debugged, and verified |
Pierre-Emmanuel Gaillardon | 0.9 | Univeristy of Utah | Open-source FPGA generation framework |
Sheief Reda | 0.6 | Brown University | Open-source circuits and software for thermal-, voltage-, and prcoess sensors, employing an exploratory solver based on distributed set of die sensors |
Chinh Le | 0.6 | LeWiz Communications | High-quality open-source implementations of Ethernet controllers |
In its architecture efforts, DARPA is aiming for a future where systems-on-chips contain a lot of specialized hardware—such as accelerators hardwired to do particular functions fast—instead of focusing on generalized computing. Software for these chips must still be easy to write and yet be able to take the best advantage of the accelerators involved. [For more, see last week's interview with Chappell.] The two newly funded programs under the architecture effort are called Software Defined Hardware (SDH) and Domain-Specific System on Chip (DSSoC).
According to DARPA, the SDH program aims to develop hardware and software that can be reconfigured in real time based on the sort of data being processed, adapting the computing architecture for the workload in milliseconds. Here, the big winner was a $22.7 million project headed by Stephen Keckler at Nvidia but also including researchers from MIT, the University of Illinois Urbana-Champaign, and the University of California Davis.
Principal | Funds (millions of US$) | Institutions | Description |
Stephen Keckler | 22.7 | Nvidia, MIT, University of Illinois Urbana-Champaign, University of California Davis | Reconfigurable compute arrays with explicit data orchestration |
Michael Bedford Taylor | 9 | University of Washington, Cornell University, Dini Group, Xilinx | Polymorphic hybrid ASIC/FPGA architectures |
Ron Dreslinski | 9 | University of Michigan, ARM, Arizona State University, University of Edinburgh | Fast, low energy interconnect for reconfigurable compute arrays |
Kunle Olukotun | 8 | Stanford University, SambaNova Systems | Hybrid configurable arrays coupled with on-the-fly compiler optimization |
Margaret Martonosi | 5.8 | Princeton University, Columbia University | Specialized and decoupled data compute and data supply programming for malleable tile-based computation |
Brad Gaynor | 5.5 | Systems & Technology Research, Northeaster University, Purdue University | Data dependent Just-in-Time hardware/software |
Joshua Fryman | 4.5 | Intel | High-throughput compute near memory and interconnect networks |
vivek Sarkar | 4.5 | Georgia Tech, Univeristy of Illinois Urbana-Champaign, University of Michigan, University of Southern California | Automatically discover and generate parameterized kernels for hardware and software co-optimization |
Shekhar Borkar | 2 | Qualcomm | Dynamically optimizable data path for multi-core systems |
The DSSoC arm of the architecture effort is somewhat smaller but is still funding two groups at more than $10 million each. “It is critical for the DoD to have flexible, adaptable radio systems that are capable of managing and combating a complex signal environment,” Tom Rondeau, the program manager leading DSSoC said in a press release. “These devices must be programmable like general purpose processors, but also capable of crunching a lot of math with low power.”
Principal | Funds (millions of US$) | Institutions | Description |
Daniel Bliss | 17.4 | Arizona State University, University of Michigan, University of Arizona, Carnegie Mellon, General Dynamics, ARM, EpiSyS | Heterogeneous processors with specialization to support multifunction RF systems that require dramatically increased processing flexibility at low power |
Pradip Bose | 14.7 | IBM, Columbia, Harvard, University of Illinois at Urbana-Champaign | Quickly design and implement an easily programmed domain-specific SoC for real-time cognitive decision engines within smart connected vehicles |
Mark Horowitz | 6.4 | Stanford University, SambaNova Systems | Visual computing program will use agile design techniques to enable application experts to develop ideas quickly for an optimized SoC for visual computing |
Dr. Jeffrey Vetter | 6 | Oak Ridge National Lab | New tools for performance modeling and quantitative analysis of domains to enable co-design of system-on-chip specialization |
In the materials and integration arm of the $1.5 billion effort, DARPA introduced the wining teams in two new programs, Foundations Required for Novel Compute (FRANC) and Three Dimensional Monolithic System-on-a-Chip (3DSoC). 3DSoC aims to gain a 50-fold power computation-time improvement and power reductions by growing multiple layers of interconnected circuitry atop a CMOS base. A group led by MIT’s Max Shulaker gained the lion’s share of the research funds ($61 million) for an ambitious project that aims to show that a monolithic 3D system using 90-nanometer manufacturing technology can match the performance of a CMOS chip made with today’s most advanced process technology.
Principal | Funds (millions of US$) | Institutions | Description |
Max Shulaker | 61 | MIT, Skywater, Stanford University, Carbonics | Refined RRAM and CNFET processes for commercial fabrication of an integrated, monlithic 3D SoC |
Sung Kyu Lim | 3.1 | Georgia Tech, Duke University | Optimized EDA software for complex, high-performance monlithic 3D SoC |
FRANC is focused on using new nonvolatile memory devices to perform processing in memory. Naresh Shanbhag from the University of Illinois is leading the biggest project in that category, exploring in-memory processing with embedded MRAM. But others will look at more exotic types of memory. “FRANC seeks to utilize new materials and devices to make 10x advances in embedded, non-volatile memories with the speed of static random access memory (SRAM) and the density of storage-class memory,” said Y.K. Chen, the program manager leading FRANC, in a press release. “These advances could allow emerging memory-centric computing architectures to overcome the memory bottleneck presented in current von Neumann computing.”
Principal | Funds (millions of US$) | Institutions | Description |
Naresh Shanbhag | 8.3 | University of Illinois Urbana-Champaign | Energy-efficient in-memory computing with non-volatile memory |
David Thompson | 6.7 | Applied mateials, ARM, Symetrix, University of Colorado, University of British Columbia | New correlated electronic multi-level memory devices with robust Mott transition |
Wei Yi | 3.4 | HRL Laboratories | New active and passive memristor devices for neuromorphic functios and computation |
Noah Sturcken | 3.1 | Ferric | Integrated magnetic devices for efficient power devices |
Sudhakar Pamarti | 1.9 | UCLA | Energy-efficient spintronics-based stochastic computing system with non-volatile magnetic memory |
Jian-Ping Wang | 0.8 | University of Minnesota, University of Arizona | High-density magnetic memory devices with synthetic antiferromagnetic free layers for fast switching speed with low voltage and energy |
Samuel K. Moore is the senior editor at IEEE Spectrum in charge of semiconductors coverage. An IEEE member, he has a bachelor's degree in biomedical engineering from Brown University and a master's degree in journalism from New York University.