SBIR-STTR Award

Extensible Streaming RISC-V DSP Processor Architecture
Award last edited on: 9/6/22

Sponsored Program
SBIR
Awarding Agency
DOD : Army
Total Award Amount
$111,447
Award Phase
1
Solicitation Topic Code
A20-003
Principal Investigator
David Richie

Company Information

Brown Deer Technology LLC

1641 Denwright Court
Forest Hill, MD 21050
Location: Single
Congr. District: 01
County: Harford

Phase I

Contract Number: W58RGZ-21-C-0024
Start Date: 5/28/20    Completed: 4/21/22
Phase I year
2020
Phase I Amount
$111,447
The DARPA-funded RISC-V architecture reflects a potentially disruptive shift towards open-hardware design and standardization in a computer industry that has been dominated by closed proprietary instruction set architectures (ISAs) for decades. The modular and extensible design of RISC-V enables a foundation for generally-specialized architectures for greater power-performance efficiency as we witness an end to Dennard Scaling and Moore's Law. The potential for a RISC-V based Digital Signal Processor (DSP) is significant and would address many critical challenges in the deployment of signal processing applications. The engineered extensibility in the RISC-V ISA provides additional support for exploring a RISC-V DSP with both standards-based extensions and proprietary application-specific customization. We propose to develop a family of customizable RISC-V DSP architectures targeting different application deployments ranging from ultra-low-power embedded to more high-performance general-purpose processors with DSP specialization. Using a common foundation of interoperable DSP ISA extensions, special attention will be given to the more challenging constraints of embedded edge-device applications with strict power and space constraints. Each DSP processor will implement the base RISC-V ISA along with complementary standard extensions appropriate for the target application space. A modular DSP extension to the RISC‑V ISA will be developed to include extensions for complex floating-point arithmetic, real and complex fixed-point arithmetic, and streaming instructions. All processors will enforce a strict true Harvard architecture memory model that goes even further to include separable memory segments for application performance. Enforcing a true Harvard architecture for DSP applications has important implications for performance security

Phase II

Contract Number: ----------
Start Date: 00/00/00    Completed: 00/00/00
Phase II year
----
Phase II Amount
----