CNN Sensor Analytics With Hybrid-Float6 Quantization on Low-Power Embedded FPGAs
نویسندگان
چکیده
The use of artificial intelligence (AI) in sensor analytics is entering a new era based on the ubiquitous embedded connected devices. This transformation requires adoption design techniques that reconcile accurate results with sustainable system architectures. As such, improving efficiency AI hardware engines as well backward compatibility must be considered. In this paper, we present Hybrid-Float6 (HF6) quantization and its dedicated design. We propose an optimized multiply-accumulate (MAC) by reducing mantissa multiplication to multiplexor-adder operation. exploit intrinsic error tolerance neural networks further reduce approximation. To preserve model accuracy, quantization-aware training (QAT) method, which some cases improves accuracy. demonstrate concept 2D convolution layers. lightweight tensor processor (TP) implementing pipelined vector dot-product. For portability, 6-bit floating-point (FP) wrapped standard FP format, automatically extracted proposed hardware. hardware/software architecture compatible TensorFlow (TF) Lite. evaluate applicability our approach CNN-regression for anomaly localization structural health monitoring (SHM) application acoustic emission (AE). framework demonstrated XC7Z007S smallest Zynq-7000 SoC. implementation achieves peak power run-time acceleration 5.7 GFLOPS/s/W $48.3\times $ , respectively.
منابع مشابه
HYREP: A Hybrid Low-Power Protocol for Wireless Sensor Networks
In this paper, a new hybrid routing protocol is presented for low power Wireless Sensor Networks (WSNs). The new system uses an integrated piezoelectric energy harvester to increase the network lifetime. Power dissipation is one of the most important factors affecting lifetime of a WSN. An innovative cluster head selection technique using Cuckoo optimization algorithm has been used in the desig...
متن کاملImplementing Efficient Low-Power PCIe Interfaces with Low-Cost FPGAs
A history of architectural and process advancements has enabled Altera® Cyclone® V FPGAs to be used in numerous low-cost and low-power applications in the industrial, automotive, military, communication and consumer markets, among others. This white paper outlines a real-life PCI Express® (PCIe®) Gen1x4 reference design including a DDR3 memory controller. It shows just how effective Cyclone V F...
متن کاملAugmenting Fpgas with Embedded Networks-on-chip
FPGAs are increasing in capacity, allowing the implementation of ever-larger systems with correspondingly increasing bandwidth demands. Additionally, modern FPGAs are becoming a heterogeneous platform that includes many fast dedicated blocks such as processor cores, and high-speed I/Os such as DDR3 memory and PCIe. It is becoming a challenge to connect systems in these large heterogeneous FPGAs...
متن کاملA Hybrid Memory Architecture for Low Power Embedded System Design
On-chip memories are one of the most power hungry components of today’s system on a chips (SoCs). The on-chip memories generally use higher supply (Vdd) and threshold (Vth) voltages than those of logic parts to suppress the static power consumption without increasing the access delay of the memories. This design policy, however, increases the dynamic power consumption since the dynamic power co...
متن کاملReducing Total System Cost with Low-Power 28 nm FPGAs
When building systems for high-volume applications, it is very important to keep costs in check. There are several dimensions that affect total cost of ownership beyond the price per part. These include the power demands of the silicon, total bill of materials (BOM) cost, and the productivity of the engineers who design and test the system. It is important to choose an FPGA vendor that has cons...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: IEEE Access
سال: 2023
ISSN: ['2169-3536']
DOI: https://doi.org/10.1109/access.2023.3235866