Volet 1: Mon projet en 180·106 μs – Chaque candidat va faire connaître son projet via une courte présentation orale de 3 minutes.
Volet 2: Mon project en action – Démonstration technique de microsystèmes en direct.
Projets en compétition
(cliquez sur le titre pour voir une description complète)
Niveau cycles supérieurs
P1 – ECG beat analysis using discrete wavelet coefficient
par Azfar Adib (Ph.D.), Concordia
P2 – Wireless headstage for dual band electrophysiology and optogenetic stimulation
par Guillaume Bilodeau (Ph.D.), Université Laval
P3 – Bacteria Energy Recovery System Using Natural Soil Bacteria in a Microbial Fuel Cells
par Nathaniel Brochu (M.Sc.), Université Laval
P4 – Switched input reset for drift control of a contactless ECG electrode
par Marc-Alexandre Chan (Ph.D.), Concordia University
P5 – High speed rotary MEMS micromotor for OCT application
par Amit Gour (Ph.D.), École de technologie supérieure
P6 – Speech Enhancement Using Complementary Neural Networks
par Mojtaba Hasannezhad (Ph.D.), Concordia University
P7 – Flexible Hybrid Electronic: printed capacitive sensor for ECG acquisition
par Mathieu Lessard-Tremblay (M.Sc.), École de technologie supérieure
P8 – Spectromètre à Haute Résolution à Base de Nanoparticules d’Or pour la Détection de Neurotransmetteurs
par Shimwe Dominique Niyonambaza (Ph.D.), Université Laval
P9 – Multibans RF Energy Harvesting for IoT & Biomedical Applications
par Seyed Mohammad Noghabaei (Ph.D.), Polytechnique Montréal
P10 – The EcoChip: A Wireless Multisensor Platform For Comprehensive Environmental Monitoring
par Karim Ouazza (M.Sc.), Université Laval
P11 – Highly Integrated MEMS resonator for gas concentration sensor
par Alberto Prudhomme (M.Sc.), École de technologie supérieure
P12 – 12.5 Gb/s 1.93 pJ/bit Optical Receiver Exploiting Silicon Photonic Delay Lines for Clock Phases Generation Replacement
par Bahaa Radi (Ph.D.), McGill University
P13 – Hand Gesture Recognition with a Custom High-Density Surface
par Simon Tam (M.Sc.), Université Laval
P14 – A Deep Neural Network Based Kalman Filter for Time Domain Speech Enhancement
par Hongjiang Yu (Ph.D.), Concordia University
This work aimed to develop a new approach to identify, different types of electrocardiogram (ECG) beats using discrete wavelet transform (DWT) coefficients. Five types of cardiac phenomena were considered and for each of those some particular records from MIT-BIH arrhythmia database were selected. For these records DWT coefficients up to level 4 were calculated in the MatLAB7.40 environment, using different types of mother wavelets. A comparison was made between the performances of different types of mother wavelets to select the mother wavelet providing the best result.
This project is the development of a wireless optogenetic headstage which allows electrophysiological recording of both neural action potential (AP) and local field potentials (LFP) simultaneously with multichannel optical stimulation. The device is designed for use in experiments with freely moving rodents, like small laboratory mice. It can record up to 32 channels of electrophysiological activity and stimulate optically with up to 4 channels using only commercial off the shelf (COTS) components. An embedded digital field-programmable gate array (FPGA) is used within the system to allow real time digital separation of AP and LFP features. It also performs AP detection and compression for data reduction allowing wireless acquisition of all channels. The digital filters are optimized to reduce latency, so the 32 channels can be time multiplexed on a single data path to reduce resource utilization. The signal separation allowing parallel processing of the AP and LFP to greatly reduce the amount of data to transmit and allow a higher channel count for both signals.
In the last decade wireless connections was significantly boosted thanks to telephones, computers, cars, among many others. This concept is based in the Internet of Things (IoT). To push technology towards the development of wire free power, wireless power supplies are currently being engineered. Soil Microbial Fuel Cell (MFC) is one among these types of energies. Soil MFCs are a bioelectrochemical system that can drive an electric current using bacteria. They contains two electrodes; cathode and anode. Currently, soil MFCs can only produce minimal amount of voltage and current. However, the growing need for sustainable green energy all over the world and reducing the dependency on fossil energy in the addition to the development of ultra-low power electrical components facilitate the integration of MFCs. We created a full autonomous Bacteria Energy Recovery System (BERS) using soil Microbial Fuel Cell (MFC). We attained voltage up to 7 V and 120 μA using 7 MFCs thanks to a two-cycles based power harvesting module. Furthermore, we successfully powered a newly designed, ultra-low power, portable pH probe using the power previously harvested. The precision on pH probe is 0.1 by unit of pH.
Electrocardiography (ECG) is an essential cardiological diagnostic tool, but it has a number of limitations: it uses disposable electrodes, and requires a specialist to precisely place them. In recent years, contactless electrocardiogram (cECG) electrodes have been developed that can read ECG signals at a few millimetres distance (on skin or through thin clothing), to improve flexibility for quick screening or long-term monitoring. Some of the principal remaining challenges of cECG, however, are motion artefacts and triboelectric charge transfer. This causes noise and a slow drift of the signal. We designed a cECG sensor containing a custom integrated circuit amplifier with ultra high input impedance and implemented a FET switch input biasing circuit. This switch, when closed, allows quick discharging of any accumulated charge that has caused the output to drift, but maintains the required input impedance for normal operation when open; this allows the system to more quickly obtain stable readings, for example, when a patient initially lies down on a sensor pad, or if the signal drifts during data acquisition. For this demonstration, we show a live capture of a simple two-lead ECG using two sensor units, with the input switch manually controlled.
We present the design and implementation of a chip-scale high speed rotary micromotor for the miniaturization of a rotary polygon scanner to be used in optical coherence tomography (OCT) system. Micromotor optimization is performed using finite element analysis. Single rotor/stator pair is simulated to extract the driving torque at different voltage for multiple stator/rotor overlap angle. The various design parameters of the micromotor such as radius of the rotor, gap between the stator and the rotor, rotor/stator angular width of the rotor are deduced analytically. The maximum rotational speed of the micromotor is approximately 8000 rpm for a square wave excitation with peak-to-peak voltage of 200V and operating frequency up to 1050 Hz has been achieved. Based on our findings, we propose a new architecture for electrostatic micromotor to attain higher speed and scan rates.
Nowadays, machine learning methods have substantially advanced benefiting from the exponential growth of big data and increasing computational resources. Many human-machine interfaces based on speech using automatic speech recognition (ASR) for interaction with intelligent electronic devices, such as dictation systems, voice-enabled search for mobile devices and voice-controlled home entertainment systems. As the human-machine interfaces, say Google Home and Amazon Echo, has to perform under difficult acoustic conditions, like in a living room involves with a wide variety of highly non-stationary noise sources, such as children’s voices, television or ambient music, robustness to noise gets an increasingly important issue which becomes worse by room reverberation. Robustness to noise and reverberation remains a challenging problem that is actively driving research on speech enhancement (SE). In this project, we investigate new SE methods based on deep neural networks (DNN) for ASR applications. Supervised SE algorithms can be broadly divided into three main components of learning machines, training targets, and acoustic features. Our project is focused on these three consisting of investigating the best feature set so as to represent the noisy and reverberant speech, designing a hybrid DNN core engine specifically tailored to the SE task, and adopting an appropriate and inclusive training target.
We developed an electrode that acquire ECG through clothing or without any requirements in regard of skin preparation. The electrode is flexible and can be manufactured using standard techniques such as screen printing and pick and place in term of active component assembly. The sensing surface of the electrode is made from ink doped with silver nanoparticles and is insulated from the patient using a laminated dielectric. The substrate of print is a sheet of ST505, and use vias to transfer the signal from the sensing surface to the first stage of amplifier. The components are assembled on a classical PCB and is then bonded to the bottom of the substrate of print. Miniaturization of the active components is a key milestone to achieve to reduce the footprint of the rigid part of the assembly. On long term vision, only a little IC would act as the first stages of information preparation before relying on substages to treat and filter the data, on devices that are external to the sensors.
Les neurotransmetteurs sont parmi les molécules les plus difficiles à détecter à cause de leur très faible concentration dans les liquides physiologiques. Les méthodes existant pour leur détection nécessitent des manipulations complexes et de gros appareils qui utilisent trop d’énergie pour être portable. Ce travail présente le prototype d’un spectromètre compact conçu pour la spectroscopie visible des nanoparticules d’or sphériques fonctionnalisées pour la détection de la dopamine. Le système utilise des nanoparticules d’or ultrastables ce qui permet la détection de la dopamine sans cause d’agrégation dans le système de détection. Permettant ainsi la détection en continu de la dopamine et le recyclage des nanoparticules d’or.
RF energy harvesting (RF-EH) is a common approach for the Internet of things (IoT) and self-power micro-systems such as wireless sensors, wearable devices and biomedical implants due to availability of signals everywhere. The main challenge for RF power harvester design is the sensitivity, which is the minimum available RF power required to guarantee the output power at a specific level. Multiband RF-EH design not only increases the system availability but also increases sensitivity when more than one frequency band is available, as long as the power from different bands is combined at the output. We demonstrate a new RF-DC power converter with both dynamic and static self-compensating schemes to reduce the threshold voltage of rectifying devices. The proposed scheme overcomes challenges for achieving high power conversion efficiency (PCE) at ultra-low input power. Moreover, a new power management unit (PMU) is designed to increase system availability performing harvesting from multi-band frequencies.
The EcoChip is a new autonomous wireless sensor platform intended for culturing and monitoring the growth of microorganisms and their environmental conditions in situ, in harsh environments, such as in northern climates. This platform includes a layered multiwell plate that allows the growth of single strain microorganisms, within a well of the plate, isolated from environmental samples from northern habitats. It can be deployed in the field for continuous monitoring of microbial growth within 96 individual wells through a multichannel electro-chemical impedance (EIS) monitoring circuit. The EIS monitoring system uses highperformance off-the-shelf electronic components, presents low excitation voltage signal not to harm the cells and has a calibration network for high precision. Additional sensors are provided for measuring environmental parameters such as luminosity, humidity, and temperature. The embedded electronic board is equipped with flash memory to store sensor data over long periods of time, as well as with a low-power micro-controller, and a power management unit to control and supply all electronic building blocks. When a receiver is located within the transmission range of the EcoChip, a low-power wireless transceiver allows transmission of sensor data stored in the flash memory. The performance of the system was successfully measured in vitro in a
Measuring concentration of gases on air has been a subject of research both academically and industrially due to its wide range of applications. This project applies the principle of resonance in silicon micro-structures and the shift of the natural resonance in proportion of the addition of mass in the resonator. The initial gas to be measured is CO2 with concentration ranges from 250 ppm to 10,000 ppm. To be able to obtain a reliable measurement, different capture coating materials are evaluated, Polyethylenimine, graphene and PEI/GO thin film. The application of the coating over the resonator surface represents a challenge because the size of the structures can reach 20μm. Different techniques are evaluated, as the design of an ultrasonic micro-probe that can be used for the deposition of the micro-drop in the surface with the help of an electric charge that settles the drop in the correct position. For the performance evaluation of the geometries a first chip has been produced using electrostatic resonators and a second chip was produced using piezoelectric resonators. The closed loop controller was made by the integration of a sustaining amplifier and all the system was packed to obtain a fully operative resonant sensor.
This project is about a high-speed optoelectronic receiver implemented in 65 nm CMOS technology. The receiver utilizes only two clock phases instead of the four conventionally used in a quarter-rate clocking system. This two-clock phase system is enabled by a passive silicon photonic split and delay structure that eliminates the need for a quadrature clock phase generator and all the associated buffers. Moreover, the outputs of the receiver are demultiplexed which further helps reducing power consumption in the digital part of the system. The receiver also employs inter-stage AC coupling and is mounted on a high-speed printed circuit board (PCB). The impact of AC coupling and PCB parasitics is investigated. The functionality of the receiver is validated by high-speed optical measurements. The receiver achieves an error-free transmission (BER < 10-12) up to a data rate of 12.5 Gb/s with an energy efficiency of 1.93 pJ/bit and sensitivity of -4 dBm from a 1 V supply.
This work presents a real-time fine gesture recognition system for multi-articulating hand prosthesis control, using an embedded convolutional neural network (CNN) to classify hand-muscle contractions sensed at the forearm. The sensor consists in a custom non-intrusive, compact, and easy-to-install 32-channel high-density surface electromyography (HDsEMG) electrode array, built on a flexible printed circuit board (PCB) to allow wrapping around the forearm. The sensor provides a low-noise digitization interface with wireless data transmission through an industrial, scientific and medical (ISM) radio link. An original frequency-time-space cross-domain preprocessing method is proposed to enhance gesture-specific data homogeneity and generate reliable muscle activation maps, leading to 98.15% accuracy when using a majority vote over 5 subsequent inferences by the proposed CNN. The obtained real-time gesture recognition, within 100 to 200 ms, and CNN properties show reliable and promising results to improve on the state-of-the-art of commercial hand prostheses. Moreover, edge computing using a specialized embedded artificial intelligence (AI) platform ensures reliable, secure and low latency real-time operation as well as quick and easy access to training, fine-tuning and calibration of the neural network. Co-design of the signal processing, AI algorithms and sensing hardware ensures a reliable and power-efficient embedded gesture recognition system.
Speech enhancement (SE) has been extensively applied in a wide range of fields such as speech recognition, wireless communications, hearing aids and smart home devices, where the received input speech signals are often corrupted by different kinds of noises. The main purpose of SE is to improve speech quality and intelligibility, so as to obtain better user experience in those applications. In this study, we present a novel deep neural network (DNN) based Kalman filter (KF) algorithm for speech enhancement, where DNN is applied for estimating key parameters in the KF, namely, the linear prediction coefficients (LPCs). Our method outperforms the existing KF based speech enhancement
methods in terms of both speech quality and intelligibility.