Eric
Wu.
M.Eng. in EECS [at] University of California, Berkeley
Welcome, I'm Eric Wu, currently studying M.Eng. for IC Design in EECS at UC Berkeley. I'm interested in hardware engineering, especially computer architecture and other VLSI-related projects. I was a hardware validation intern at Inteo Corp. in Taiwan, developing circuit automation tools. Further, I was hired as a digital IC design intern at MediaTak Inc. in Singapore, facilitating CPU verification process. Also, I'm a dreamer, and an innovative troublemaker. So, in this site, I want to share my special experiences with you. Please immerse in here, deeply, and enjoy it.
Awards
Outstanding Performance Scholarship (Cash Prize 2600 USD)
Honored of academic and research excellence out of 7,000 students in NTU. Below is the picture taken with Chung-Ming Kuan Chancellor at NTU.
Fisrt Place in Synopsys ARC AIoT Design Contest (Cash Prize 2600 USD)
iRobot is trying to enhance dirver's awarenessby by utilizing image segmentation. We can segment the vehicles in front of the current driver and warn if necessary. However, ultimately, we want to demonstrate that ARC processor can achieve scene understanding to do much more insteresting application in the future. For example, smart movement robot like household robot cleaner.
Second Place in Arm Design Contest (Cash Prize 3300 USD)
We want to help hearing disorder people to make sure their fundamental life requirements by using what we learnt. Therefore, we developed a gesture regonition algorithm on Arm emebedded system using only one 9-DOF IMU to achieve real-time sign language translation . In the end, we awarded with second place in Arm Design Contest with cash prize 3300 USD.
Second Place in Undergraduate Innovation Award Among 20+ Teams
To fulfill my inner nature of innovative, in a lab of designing a chip, we designed and developed a LED cubic controller chip which showcased stunning visual effect by using "real hardware." First, We wired up a 8x8x8 LED cube with circuit schematics. And later, we integrated it with FPGA-prototyping and verified by RTL codes. Finally, based on our FPGA design, we successfully tapped out a chip based on Design Compiler and Innovus.
Selected Projects
Distributed Training on Multi-GPUs Platforms
We aimed to explore architecture and system-level of machine learning models. We surveyed issues about how to train large ML models. Based on PipeDram, we were able to devise a efficient partitioning algorithm that made sure load-balancing among workers Also, we leveraged GPipe's efficient pipeline flow with the partitioning algorithm to gurantee better accuracy results and larger throughput per epoch. In the end, we achieved up to 133% throughput improvement on large ML models (ex. Vgg19, ResNet)
EL-Wire Controlled Suits
Beside academy, I also devoted myself into other interests, like music and dancing. So, I want to combine them with my engineering skills. Thus, I directed on this project, helped integrate electroluminescentwire with Linkit embedded system to build a controller CPU. In the end, we showcased our work in dancing performance, and "lighten up the stage with codes."
CNN FPGA Accelerator
We learnt current trend of CNN machine learning algorithm. But, as a hardware student, we were not enough with that. We utilized our talent to accelerate the operations by implementing a CNN architecture on FPGA. Finally, from RTL to FPGA verification, we familiarized ourselves with CNN opertations and achieved real-time computing.
Suits Recommendation System
To Improve daily-life experiences through AI-driven technologies, we leveraged speech and image recognition on RPI embedded system to help people choose the best fitting suits simply by speaking out their need on the phone, and suits, boom, would pop out.
Skills
- C/C++
- Verilog
- System Verilog
- Python
- Chisel(Scala)
- Assembly
- NC-Verilog
- Design Compiler
- Innovus
- Verdi
- Virtuoso
- Quartus II (Altera DE2-115)
- ZSim
- Intel Pin
- Gem5
- QEMU
- Spike
- riscv-tools
- Raspberry Pi
- Arm
- ARC
- Linkit