A Low-Power Transprecision Floating-Point Cluster for Efficient Near-Sensor Data Analytics

Abstract

Recent applications in low-power (1-20 mW) near-sensor computing require the adoption of floating-point arithmetic to reconcile high precision results with a wide dynamic range. In this paper, we propose a low-power multi-core computing cluster that leverages the fined-grained tunable principles of transprecision computing to provide support to near-sensor applications at a minimum power budget. Our solution -based on the open-source RISC-V architecture -combines parallelization and sub-word vectorization with a dedicated interconnect design capable of sharing floating-point units (FPUs) among the cores. On top of this architecture, we provide a full-fledged software stack support, including a parallel low-level runtime, a compilation toolchain, and a high-level programming model, with the aim to support the development of end-to-end applications. We performed an exhaustive exploration of the design space of the transprecision cluster on a cycle-accurate FPGA emulator, varying the number of cores and FPUs to maximize performance. Orthogonally, we performed a vertical exploration to identify the most efficient solutions in terms of non-functional requirements (operating frequency, power, and area). We conducted an experimental assessment on a set of benchmarks representative of the near-sensor processing domain, complementing the timing results with a post place-&-route analysis of the power consumption. A comparison with the state-of-the-art shows that our solution outperforms the competitors in energy efficiency, reaching a peak of 97 Gflop/s/W on single-precision scalars and 162 Gflop/s/W on half-precision vectors. Finally, a real-life use case demonstrates the effectiveness of our approach in fulfilling accuracy constraints.

ControlPULP: A RISC-V On-Chip Parallel Power Controller for Many-Core HPC Processors with FPGA-Based Hardware-In-The-Loop Power and Thermal Emulation

“…In our instantiation, we use one internal pipeline stage, which is sufficient to meet our frequency target. Furthermore, Montagna et al [30] show that this configuration achieves high performance and reasonable area/energy efficiency on many benchmarks.…”

ControlPULP: A RISC-V Power Controller for HPC Processors with Parallel Control-Law Computation Acceleration

“…In our instantiation, we use one internal pipeline stage, which is sufficient to meet our frequency target. Furthermore, Montagna et al [15] show that this configuration achieves high performance and reasonable area/energy efficiency on a large number of benchmarks.…”

Symmetry in the Bit-Stream Converter Design

“…The development of cyber-physical systems is due not only to new technologies of data transmission, but also to the possibility of creating integrated elements that simultaneously perform the following functions: primary processing of information about a physical object or phenomenon of various natures into an electrical signal; intermediate transformation to obtain the necessary form of data representation; and computational conversions providing the formation of the output signal transmitted to the core located on the next level of cyber-physical architecture. Due to this integration, part of the computational processing is moved from the processing core closer to the sensing element, realizing the computation and analytics "next to the sensor" paradigm [2,3]. Within the "computation near the sensor" paradigm, there is a need to develop special devices that will provide computation near the sensors.…”