Behavioural synthesis of SGD using the CCC framework: a simple XOR-solving MLP

Behavioural synthesis enables the automation of the design process by generating task-specific hardware configured for either FPGA and SoC platforms or custom silicon devices such as ASICs. Relevant commercial tools’ flows can bring significant benefits for software developers with no hardware design expertise. Our Custom Coprocessor Compilations (CCC) high level synthesis tool is leveraged in this work to synthesize a FPGA design for stochastic gradient descent (SGD), a cornerstone optimization approach into today’s modern deep neural networks. A simple 3-input-XOR-solving, multilayer perceptron (MLP) is implemented and transformed into a Register Transfer Level (RTL) VHDL hardware microarchitecture using the CCC hardware synthesizer. The produced VHDL is subsequently verified for correct functionality in GNU Ada. Results validate our motivation for accelerated performance, targeted to low-powered, autonomous devices.

     

The Spider DMA: A miniature radial differential mobility analyzer

Abstract

The Spider differential mobility analyzer (DMA) is a novel, miniaturized radial DMA developed to provide size classification in the 10–500?nm range for applications requiring high portability and time resolution. Its external dimensions are ～12?cm in diameter by 6?cm in height (excluding tubing); it weighs ～350?g, and is designed to operate at 0.6–1.5?L/min sheath and 0.3?L/min sample flowrates. It features a new sample inlet geometry that is designed to produce a uniform azimuthal particle distribution at the entrance of the classifier, optimized sample/sheath flow streams introduction in the classifier to minimize particle delays, and extension of the electric field interaction volume for ～30% enhanced dynamic range. Based on three-dimensional finite element simulations of flows, electric fields, and particle trajectories, we demonstrate that the Spider DMA transfer functions can be predicted with high fidelity using a parameterized fit based on the Stolzenburg semi-analytical model. Experimental characterization of the instrument response with size-selected particles confirmed close agreement with model prediction; mobility size response is linear over three orders of magnitude in mobility span. Electrical ground shielding of the external surfaces of the DMA has been found to be necessary to avoid particle losses associated with field effects as the high voltage operating limit is approached. The mean deviation between the reference size of polystyrene latex spheres and the Spider DMA measurement is less than 2%, corroborating its high sizing precision and potential for high quality size distribution measurements.

Copyright © 2019 American Association for Aerosol Research     

Comparative performance of a thermal denuder and a catalytic stripper in sampling laboratory and marine exhaust aerosols

The performance of a thermal denuder (thermodenuder—TD) and a fresh catalytic stripper (CS) was assessed by sampling laboratory aerosol, produced by different combinations of sulfuric acid, octacosane, and soot particles, and marine exhaust aerosol produced by a medium-speed marine engine using high sulfur fuels. The intention was to study the efficiency in separating non-volatile particles. No particles could be detected downstream of either device when challenged with neat octacosane particles at high concentration. Both laboratory and marine exhaust aerosol measurements showed that sub-23 nm semi-volatile particles are formed downstream of the thermodenuder when upstream sulfuric acid approached 100 ppbv. Charge measurements revealed that these are formed by re-nucleation rather than incomplete evaporation of upstream aerosol. Sufficient dilution to control upstream sulfates concentration and moderate TD operation temperature (250°C) are both required to eliminate their formation. Use of the CS following an evaporation tube seemed to eliminate the risk for particle re-nucleation, even at a ten-fold higher concentration of semi-volatiles than in case of the TD. Particles detected downstream of the CS due to incomplete evaporation of sulfuric acid and octacosane aerosol, did not exceed 0.01% of upstream concentration. Despite the superior performance of CS in separating non-volatile particles, the TD may still be useful in cases where increased sensitivity over the traditional evaporation tube method is needed and where high sulfur exhaust concentration may fast deplete the catalytic stripper adsorption capacity.

Copyright © 2018 American Association for Aerosol Research