Hello, I'm Jack!
I'm an ECE PhD Student.

in The Embodied System Lab at Northwestern
Jack Thoene
Jack Thoene

I research power-constrained computing systems with an emphasis on co-design across materials, hardware, and software. My work is grounded in the principle that transformational advances emerge when the entire stack—from silicon and power electronics to system software and user interfaces—is optimized for a specific application. My long-term focus is on translating advanced capabilities, including hyperspectral imaging and semiconductor fabrication, into accessible platforms that enable the next generation of engineers to be multidisciplinary from the outset.

I view people as an organization's most critical asset. As the senior student in the VAK Embodied Systems Lab, I help manage, equip, and mentor an interdisciplinary group of ~30 students spanning Electrical and Computer Engineering, Computer Science, Mechanical Engineering, Environmental Engineering, Materials Science, Applied Mathematics, Physics, and Chemistry. I emphasize team formation, technical integration, and mentorship as essential system components:

People are the pillar of full-spectrum engineering.

View my work arrow-right
Download CV

Publications

Contact

LinkedIn GitHub

Talk is cheap. Let's get to building. The form below will automatically send to my personal email. If you're ready to build something special, let's find the 80% solution and take off at a sprint together.

Please enter your name.
Please enter a message.
LOAM Assembly

LOAM: Low-Cost Low-Power Activity-Aware Soil Moisture Sensing Platform

Deep understanding of a field's soil moisture content is the lead- ing indicator for predicting crop yields and making data driven decisions for irrigation and application of topical chemicals for drought resilience. Despite this importance, the cost of adopting and maintaining IoT infrastructure prevents modern farms from em- ploying widespread real time soil moisture sensors. LOAM presents an end-to-end platform of buried battery-free sensor nodes and a mobile basestation that leverages the farmer's daily routine for data retrieval. Each LOAM node features a self-powered galvanic soil-moisture probe, employing a high impedance analog front end to enable durability. Operating entirely on harvested solar en- ergy for up to 21 days on a single capacitor charge, each node collects soil moisture, temperature, and environment condition data. Using a predictable finite-state machine, handshake-based data exchanges occur with a basestation affixed to standard farm- ing vehicles designed to listen for the nodes while moving through the farm. LOAM organizes all sensor, link-quality, and location data into an easy-to-interpret dashboard to seamlessly integrate with the farmer's everyday routine. Costing less than $35, LOAM is a financially accessible, accurate, and easily scalable platform that enables persistent, regular data collection from the most rural plots without adding to or impeding farming operations. Experimental evaluation demonstrates reliable communication over 1 km at 2 dBm transmit power, stable sensor readings over 70 days of indoor operation, and continuous data recovery during multiple periods of intermittent connection.

This work is currently under review by the publishing editors, more information will become available after publication.
Go to Publisher View on GitHub Download PDF

This work is under review by the publishing editors, links will become available after publication.

MANTIS Poster

MANTIS: Manufacturable Application-Specific Narrowband Tunable Filters for Inference-In-Sensor

We present MANTIS: Manufacturable Application-Specific Narrowband Tunable Filters for Inference-In-Sensor, a framework to design and implement application-specific filters to augment image sensing platforms with hyperspectral image- like performance without the complexity, cost, and computational overhead. MANTIS provides a methodology for creating custom hyperspectral datasets collected using low-cost pushbroom HSI systems to identify task-relevant spectral bands via machine learning. These bands are then translated into realizable nar- rowband optical filters, enabling inference-in-sensor without full hyperspectral acquisition. We demonstrate the feasibility of this approach by learning and simulating combined binary masks based on open source datasets.

This work is currently under review by the publishing editors, more information will become available after publication.
Go to Publisher View on GitHub Download PDF

This work is under review by the publishing editors, links will become available after publication.