Softcore Streaming Processors for FPGA

Lead Research Organisation: Queen's University Belfast


Modern FPGA represent the pinnacle of the embedded device market for computationally intensive applications; providing an unprecedented 1.2TMACs of fixed-point on-chip performance, and 28Tbits/s memory bandwidth, they offer massive resources with which to build embedded realisations of applications such as Digital Signal Processing (DSP), and the ability to optimise the cost of the embedded architecture in an application specific manner.They are, however, critically hamstrung by programming problems. In stark contrast to the software-based programming style of the majority of alternative devices, FPGA normally require the use of hardware design languages and techniques to derive the best implementation.This project proposes a new approach to solve this problem for FPGA DSP. Having shown that simple programmable processors can achieve real-time performance and cost similar to dedicated hardware on FPGA, we propose a project which uses these processors to build FPGA DSP architectures. We propose to develop an ultra-efficient processor architecture and a design methodology to translate a dataflow graph DSP application model into a heterogeneous network of the simple processors, tailored to the application requirements. Further, we will automate this process and apply to real world application demonstrators from our industrial partners Xilinx Inc., the FPGA device market leader and National Instruments, a premier global supplied of FPGA embedded products.

Planned Impact

What Benefits Will Accrue from This Research? We foresee benefits from this project in the international academic knowledge, economic and skills and trained people areas. Academic benefits are summarised in the academic beneficiaries statement. What are The Economic Benefits of this Research? The economic impact of this project is considerable. It address a critical problem for the entire embedded systems design industry, that of programming parallel processors. The UK has a vested interest in work of this type, as it hosts 40% of the European independent semiconductor design market, from embedded processors (ARM, Picochip) to chipsets (ICERA, Cambridge Silicon Radio). Furthermore, the resulting FPGA design tools can produce better results than current offerings and so can have a major impact in current FPGA markets whilst increasing the uptake of FPGA to expand the size of the market itself. This should have a major impact with the UK's indigenous FPGA platform vendor (e.g. Nallatech, VMETRO - now both part of Curtiss-Wright) and system design houses (e.g. QinetiQ, Selex). We foresee an impact on FPGA platform and device vendors worldwide (due to the increased sales of FPGA devices/platforms), and design tool vendors, who will have new design tools to sell to FPGA systems houses. To ensure these impacts we have incorporated strategic relations with three industrial partners into this project, including Xilinx (the FPGA device market leader), National Instruments (NI - a premier global supplier of FPGA embedded platforms). Ensuring Economic Impact The PI and CI are co-founders of CapnaDSP Ltd., an RTS spin-out commercialising recent FPGA design tool developments in the group. As Scott Fischaber indicates in his letter of support, CapnaDSP will work closely with the project team to ensure the relevance of the resulting technology to their design problems and will potentially license the resulting technology. The increased abstraction level and more productive FPGA design route this project promises have attracted Xilinx and NI (see letters of support from Xilinx's Stephen Neuendorffer and NI's Newton Petersen). Both will supply 'real-world' application demonstrators for the resulting technology, and will interact with the project team throughout it's term. Xilinx will supply equipment and software to support this project, and we plan to have researchers at NI and Xilinx's headquarters (Austin, Texas and San Jose, USA) during the latter parts of the project to ensure impact with them. In addition, the CI will also use a long-standing relationship with Prof. Ian Phillips of ARM to investigate embedded multiprocessor application of this technology via informal meeting and engagements such as the ARM External Speakers Conference and the UK Design Forum. What are The Skills/Trained People Benefits of this Research? The researchers on this project will avail of the unique industry-facing research experience of ECIT ( at Queen's University, and will develop a unique balance of technical and industry focused research skills. Furthermore, the opportunity to interact and spend time on placement with Xilinx and NI will expose the projects' researchers to world class FPGA system design researchers and facilities, a unique opportunity for them, and one which inherently builds on the research experience based in ECIT specifically, and the UK in general. These interactions form an effective economic and skills impact strategy, ensuring close contact with industry, real-world applicability of the resulting technology, and a set of researchers uniquely balanced between technological innovation and real-world system needs. They are a key, fundamental part of this project and integrated into it's nature, planning and execution. We anticipate considerable industrial and commercial interest in the results of this project as a result of these interactions.
Description Modern electronic systems, such as advanced driver assistance, 5G communications or high performance data processing, depending the ability to be able to perform complex arithmetic operations very quickly. To do this in many cases demands custom circuit architectures realised on devices such as Field Programmable Gate Arrays (FPGAs) and an associated very demanding design process. This project has shown that this is not necessary. It has shown that the use of very large scale multicore architectures, built from very fine-grained, software-programmable processors, can achieve performance as high and cost as low as custom circuits. This is a unique result, and this project is the first to have shown it. This project has developed such a processor and a series of complex custom computer architectures based upon it and demonstrated its novelty via empirical measurement.
Exploitation Route This work may be of great importance to designers of custom computing architectures on FPGA of two kinds. The first are those who wish to realise computationally complex operations with high performance, low cost and short design cycle. The second are tool developers, who could use the processor developed under this project as a target for their compilers and design tools, for automatic generation of such custom computers from a domain-specific operation description language.
Sectors Aerospace, Defence and Marine,Chemicals,Digital/Communication/Information Technologies (including Software),Electronics,Financial Services, and Management Consultancy,Manufacturing, including Industrial Biotechology,Pharmaceuticals and Medical Biotechnology

Description InvestNI Proof-of-Concept
Amount £102,429 (GBP)
Organisation Invest Northern Ireland 
Sector Public
Country United Kingdom
Start 12/2016 
End 02/2018
Description Keysight Technologies Research Gift
Amount $35,000 (USD)
Organisation Keysight Technologies 
Sector Private
Country United States
Start 11/2016