Seyed Parsa Hemmasi

In-memory computing, deep neural networks accelerator, emerging technologies, and approximate computing are all areas of interest to me. My research focuses on process in memory architectures based on non-volatile memories for use in acceleration of deep neural networks. Currently, I am pursuing this goal at Shahid Beheshti University Digital Lab. I am a M.Sc. student in Electrical Engineering at the Azad University - Science and Research Branch (SRBIAU) under supervision of Dr. Mohammad Hossein Moaiyeri and advised by Dr. Behzad Ebrahimi. I completed my B.Sc. in Electrical Engineering at Shahid Beheshti University (SBU).I am passionate about bringing science and technology into everyday life. Personal interests include history,swimming, and ping pong.

Research

Research Interests

In-memory computing
Memory design (volatile/non-volatile)
Healthcare Devices
Deep neural networks accelerator
Emerging technologies
SRAM cell design
Approximate computing

In-memory computing architecture based on non-volatile SRAM for use in Deep Neural Networks Accelerators

In-memory computing with non-volatile SRAM improves Deep Neural Networks (DNN) accelerators by addressing the von Neumann bottleneck, which is caused by the slow transfer of data between memory and processors. Instead of constantly moving data, this architecture processes data directly within the memory, reducing energy use and speeding up computations. Non-volatile SRAM also retains data without power, making it efficient for DNN tasks that involve repetitive calculations. This approach is especially useful for edge devices that need fast, low-power computing.

Projects

Image Histogram Equalizer on ZYNQ

Software/Hardware co-design | Supervisor: Prof. Nazari

Built an image histogram equalizer module using High-Level Synthesis (HLS) and implemented it on a ZedBoard.

Design of a Back-Support Active Exoskeleton

RAISeLab, Pennsylvania State University

Designed and developed a back-support active exoskeleton employing multiple IMU and EMG sensors for enhanced worker safety during heavy object handling. Utilized Dynamixel motors for efficient and controlled assistance.

Pose Estimation with CNN for Driver’s Awareness in Connected Cars

BSc. Thesis | Supervisor: Prof. Moaiyeri

Fine-tuned pose estimation convolutional neural networks (OpenPose and UniPose) with a custom dataset to classify driver’s awareness in connected cars.

Movie Recommendation System

Developed a system that recommends movies based on user ratings and similarities with other users. Programmed as a CLI app in C++.

Hand Gesture Detection

Implemented a hand gesture detection system using Google’s Mediapipe package for hand key points extraction. The ESP32-Cam captures and sends frames, which are inferred by a custom neural network model for gesture recognition. Results are sent to the ESP32 microcontroller via UDP protocol.

Genetic Algorithm for Image Generation

Developed a genetic algorithm to generate the famous Alan Turing image from random pixels, using selection, simultaneous crossover, and enhanced adaptive mutation techniques.

Web Scraping of a Product Page

Created a Python project using BeautifulSoup4 to scrape promoted products, discounts, and prices from a product page.

Washing Machine Prototype

Developed a washing machine prototype using an ATmega32 microcontroller with a PID-controlled heater, Servo motor, multiple DC motors, an LCD display, and an intuitive user interface.

Edge Detection using VHDL

Implemented a Sobel edge detector filter using VHDL. Input images were converted to binary, convolved with a Sobel kernel, and output images were processed and reconstructed with MATLAB.

PD-Controller Ball Balancer

Created a ball balancer integrating a PD-controller to balance a ping-pong ball on a plane. The ESP8266 microcontroller sends and receives data from a custom website to control the Servo motor.