A trio of computer scientists have published a paper describing an extension to the RISC-V instruction set architecture (ISA) to add instructions for ultra-low power (ULP) wireless signal processing for the Internet of Things (IoT).
The RISC-V ISA is, as its name suggests, a reduced instruction set computing (RISC) ISA - but just because it's reduced doesn't mean it can't be extended. As an open source ISA, anyone is free to develop on top of RISC-V — including building extensions which offer specialized instructions for particular tasks.
"This work presents an instruction-set extension to the open source RISC-V ISA (RV32IM) dedicated to ultra-low power (ULP) software-defined wireless IoT transceivers," computer scientists Hela Belhadj Amor, Carolynn Bernier, and Zdeněk Přikryl write in the abstract to their paper. "The custom instructions are tailored to the needs of 8/16/32-bit integer complex arithmetic typically required by quadrature modulations. The proposed extension occupies only 2 major opcodes and most instructions are designed to come at a near-zero energy cost."
The extensions may use little energy themselves, but they offer a dramatic improvement in efficiency for wireless signal processing. In testing, using both instruction-accurate and cycle-accurate models of six test benches including frequency-shift keying (FSK) demodulation, Bluetooth Low Energy, and LoRa preamble detection, the team showed cycle count improvements ranging from 19 percent to 68 percent over a baseline 32-bit RV32IM implementation.
"To guarantee ultra-low power performance, instructions are designed to come at a near-zero power cost while reducing cycle count by more than 50 percent in complex arithmetic-intensive algorithms," the team explains. "The corresponding reduction in clock frequency results in substantial energy savings confirmed by post-synthesis power simulations of the extended core.
"By proving the feasibility of ultra-low power performance even for computationally intense protocols such as maximum bandwidth LoRa preamble detection and Bluetooth Low Energy, this work paves the way to making ultra-low power software-defined radio a reality."
The team's work was published in the journal IEEE Transactions on Computers, and is available under open-access terms via the HAL archive.