Sanjay Seshan

Hanly, B.(1st), Ospina, L.(1st), Seshan, S.(1st), Paul D.J., Niroui, F., Jan. 2024, Two-dimensional MoS₂ transistors (Poster), Microsystems Annual Research Conference (MARC)

Designed and fabricated a nano-scale 2D transistor using MoS2 channels for use in developing logic gates. Work was developed as part of MIT 6.s059 and was presented at the Microsystems Annual Research Conference (MARC) in Jan. 2024.

Built digital circuits using TFTs on silicon. Paper is linked here.

Built a PPG sensor using an Arduino and 3d printed components.

Built a ripple tank using an Arduino to demonstrate wave interference patterns

Designed and implemented a PSoC (programmable system on chip) based oscilloscope, with two analog inputs and one analog output, including full frequency analysis and user customizability. Built as part of MIT 6.115 course.

I implemented a Graph-based Vector Search accelerator on a Xilinx FPGA using SystemVerilog, built using the vivado toolchain and testing software. Write-up can be found here.

Abstract—With the application of graphs in large-scale modeling, such as social network analysis and image and video segmentation, among other applications, graphs are increasingly being used to encode and find complex relationships between data for machine learning models, leading to an increased need for optimization of these models. As a result, in order to better support model-specific algorithm efficiency, there has been work to create specialized hardware accelerators focusing on aspects such as memory accessing, latency, and resource allocation. However, current accelerators for graph problems are not scalable and can only be optimized for a single application, such as graph random walks or matrix multiplication. Existing systems also run on CPU or GPUs. Following from prior accelerator work on FPGAs, we plan to implement a graph-based vector search algorithm, based on iQAN, that runs on an FPGA to produce better algorithm performance than on existing systems and can be used for more versatile applications.

Completed layout and tapeout process for a mixed-signal, CMOS-based differential amplifier in Cadence as part of the MIT 6.2080 course

Designed and implemented a pipelined, dual-core RISC-V 32-bit processor with a shared cache hierarchy in Bluespec SystemVerilog. Synthesized design to work on AWS-based FPGA

RISC-V processor built entirely in Javascript for web-based simulation. It can be used for making visualizations for computer architecture courses. See here.

Designed and tested a custom wearable wristband sensor to detect hand movements. Developed a gesture recognition system leveraging custom wearable and machine learning analysis. Paper at ArXiv.2009.13322.

The goal of this project is to create an inexpensive, lightweight, wearable assistive device that can measure hand or finger movements accurately enough to identify a range of hand gestures. One eventual application is to provide assistive technology and sign language detection for the hearing impaired. My system, called LiTe (Light-based Technology), uses optical fibers embedded into a wristband. The wrist is an optimal place for the band since the light propagation in the optical fibers is impacted even by the slight movements of the tendons in the wrist when gestures are performed. The prototype incorporates light dependent resistors to measure these light propagation changes. When creating LiTe, I considered a variety of fiber materials, light frequencies, and physical shapes to optimize the tendon movement detection so that it can be accurately correlated with different gestures. I implemented and evaluated two approaches for gesture recognition. The first uses an algorithm that combines moving averages of sensor readings with gesture sensor reading signatures to determine the current gesture. The second uses a neural network trained on a labelled set of gesture readings to recognize gestures. Using the signature-based approach, I was able to achieve a 99.8% accuracy at recognizing distinct gestures. Using the neural network the recognition accuracy was 98.8%. This shows that high accuracy is feasible using both approaches. The results indicate that this novel method of using fiber optics-based sensors is a promising first step to creating a gesture recognition system.

Developed a system that combines on-vehicle and infrastructure-based sensors to provide a more complete view of the environment; developed a custom vision recognition system, sensor fusion model, and networking protocol design for the system. Presented work at Intel International Science and Engineering Fair in Phoenix, AZ. Paper at Arxiv:2009.03458.

Studies predict that demand for autonomous vehicles will increase tenfold between 2019 and 2026. However, recent high-profile accidents have significantly impacted consumer confidence in this technology. The cause for many of these accidents can be traced back to the inability of these vehicles to correctly sense the impending danger. In response, manufacturers have been improving the already extensive on-vehicle sensor packages to ensure that the system always has access to the data necessary to ensure safe navigation. However, these sensor packages only provide a view from the vehicle's perspective and, as a result, autonomous vehicles still require frequent human intervention to ensure safety. To address this issue, I developed a system, called Horus, that combines on-vehicle and infrastructure-based sensors to provide a more complete view of the environment, including areas not visible from the vehicle. I built a small-scale experimental testbed as a proof of concept. My measurements of the impact of sensor failures showed that even short outages (1 second) at slow speeds (25 km/hr scaled velocity) prevents vehicles that rely on on-vehicle sensors from navigating properly. My experiments also showed that Horus dramatically improves driving safety and that the sensor fusion algorithm selected plays a significant role in the quality of the navigation. With just a pair of infrastructure sensors, Horus could tolerate sensors that fail 40% of the time and still navigate safely. These results are a promising first step towards safer autonomous vehicles.

Programmed a 3D interactive robot simulator that navigates a minefield

Track your Boston area public transit. Find the location of your bus or train as well as predicted arrivals/departures. link

Uses Machine learning to give live feedback on users' exercises. Built using OpenPose and Tensorflow to process live camera data on a Raspbery Pi and detail feedback. link

Developed Web-based tournament management system for FIRST events with support to manage team submissions and a judging and scoring system for teams.

Some examples include PacMan game and a model printer. Others can be found on youtube.

Each day, there are 850 water main breaks in the US. Early detection of leaks is critical to managing costs and saving water. I did a case study on my own local township and learnt that around 20% of water is unaccounted for, but this is a relatively low number as in some regions this number can reach up to 40%. Unfortunately, all the existing solutions to detect leakages in pipes require a leak to already be present, causing all the water to still be lost. Therefore, my goal was to detect the leaking water by detecting pipe deterioration before there is an actual break and water is lost. To do this, I used the acoustic properties of pipes to infer damage. youtube link, code, article