An integrated pulsed laser driver in a 0.18-μm CMOS technology has been developed for generating configurable and stable laser diode (LD) pulses for time-of-flight (TOF) applications. The duration of the output optical laser pulses can be configured from below 100 ps up to 5 ns, with 1 ps typical resolution, by a combination of a coarse and a fine delay generator. The pulse-width stability has been designed against process, supply voltage and temperature (PVT) variations. This has been carried out using a system with two phase-locked loops (PLL) that regulate the supply voltage of the chain of inverters, found in the driver and pre-driver improving by 2.5 times the stability compared to the case where a fixed supply voltage is applied. Laser current can be configured up to 500 mA and output pulses are generated at a repetition rate up to 200 MHz.
22 December 2022 in IEEE Journal on Emerging and Selected Topics in Circuits and Systems
Nowadays Internet of Things (IoT) and mobile systems use more and more Machine Learning based solutions, which implies a high computation cost with a low energy consumption. This is causing a revival of interest in unconventional hardware computing methods capable of implementing both linear and nonlinear functions with less hardware overhead than conventional fixed point and floating point alternatives. Particularly, this work proposes a novel Radial Basis Function Neural Network (RBF-NN) hardware implementation based on Stochastic Computing (SC), which applies probabilistic laws over conventional digital gates. Several complex functions design to implement RBF-NN are presented and theoretically analyzed, such as the squared Euclidean distance and the stochastic Gaussian kernel similarity function between input samples and prototypes. The efficiency and performance of the methodology is tested over well-known pattern recognition tasks, including the MNIST dataset. The results show a low-cost methodology in terms of logic resources and power, along with an inherent capability to implement complex functions in a simple way. This methodology enables the implementation of massively parallel large scale RBF-NN with relatively low hardware requirements while maintaining 96.20% accuracy, which is nearly the same for the floating point and fixed point models (96.4% and 96.25%, respectively).
Power transmission lines transfer energy between power plants and substations by means of a linear chain of towers. These towers are often situated over extensive distances, sometimes in regions that are difficult to access. Wireless sensor networks present a viable solution for monitoring these long chains of towers due to their wide coverage, ease of installation and cost-effectiveness. The proposed LoRaBUS approach implements and analyses the benefits of a linear topology using a mixture of LoRa and LoRaWAN protocols. This approach is designed to enable automatic detection of nearby nodes, optimise energy consumption and provide a prioritised transmission mode in emergency situations. On remote, hard-to-reach towers, a prototype fire protection system was implemented and tested. The results demonstrate that LoRaBUS creates a self-configurable linear topology which proves advantageous for installation processes, node maintenance and troubleshooting node failures. The discovery process collects data from a neighbourhood to construct the network and to save energy. The network’s autonomous configuration can be completed within approximately 2 min. In addition, energy consumption is effectively reduced 25% by dynamically adjusting the transmission power based on the detected channel quality and the distance to the nearest neighbour nodes.
Internet of Things (IoT) applications and mobile systems are more and more dependent on Machine Learning based solutions, thus requiring a big computational power with a low cost in terms of power consumption. This fact has revived the interest in nonconventional hardware computing methods capable to implement complex functions in a simple way in contrast with the conventional ones. This work proposes a novel hardware/software hybrid Self-Organizing Map (SOM) implementation using stochastic computing. In turn, to support this development, several stochastic block designs are presented as the squared Euclidian distance, and the Winner-Take-All (WTA) similarity check. The capabilities and performance of the methodology is tested over a well-known classification task as the Iris flower benchmark, archiving the same classification performance than the software solutions. The proposed solution presents a low-cost methodology in terms of hardware resources and power, due to its inherent capacity to implement complex functions in a simple way. This enables the methodology to implement large self-learning classifiers based on SOM with low hardware requirements.
Deploying modern neural networks on resource-constrained edge devices necessitates a series of optimizations to ready them for production. These optimizations typically involve pruning, quantization, and fixed-point conversion to compress the model size and enhance energy efficiency. While these optimizations are generally adequate for most edge devices, there exists potential for further improving the energy efficiency by leveraging special-purpose hardware and unconventional computing paradigms. In this study, we explore stochastic computing neural networks and their impact on quantization and overall performance concerning weight distributions. When arithmetic operations such as addition and multiplication are executed by stochastic computing hardware, the arithmetic error may significantly increase, leading to a diminished overall accuracy. To bridge the accuracy gap between a fixed-point model and its stochastic computing implementation, we propose a novel approximate arithmetic-aware training method. We validate the efficacy of our approach by implementing the LeNet-5 convolutional neural network on an FPGA. Our experimental results reveal a negligible accuracy degradation of merely 0.01% compared with the floating-point counterpart, while achieving a substantial 27× speedup and 33× enhancement in energy efficiency compared with other FPGA implementations. Additionally, the proposed method enhances the likelihood of selecting optimal LFSR seeds for stochastic computing systems.