Kaida (Tommy) Zhao Email: Computer Architect & Software Engineer Department of Computer Sciences University of Wisconsin-Madison

Projects

• Exploring Proximate, a multi-tile programmable accelerator  CS 758: Adv Topics in Computer Architecture: Programming Multicore Processors, Fall 2016 Implemented parallel workloads on different parts of a Proximate compute tile Investigated scalability and speedup to improve hardware design choices

• Lock Elision and Transactional Memory Predictor in Hardware  CS/ECE 757: Advanced Computer Architecture II: Parallel Computer Architecture, Spring 2016 Transactional memory can perform better than locks depending on how often data conflict occurs in the critical section for multithreaded programs Used gem5 to simulate various lock elision and transactional memory predictor

• Superscalar Instruction Fetcher based on Branch Predictor Confidence  CS/ECE 752: Advanced Computer Architecture I, Fall 2014 Researched and developed methods for extracting confidence from branch predictor Used gem5 to simulate fetching instructions from both branch paths (taken and not taken) for various issue widths and low branch predictor confidence Achieved up to 7 percent speedup by fetching based on branch predictor confidence

• Tronsisters-32, a Nintendo Entertainment System (NES) Clone  ECE 554: Digital Engineering Laboratory, Fall 2015 Designed an Instruction Set Architecture (ISA) for computation and graphics Used Verilog for FPGA programming to create a NES Clone and support the ISA Used assembly language in the ISA to write Tron, DDR, and Pac-Man

• WISC-SP13, a pipelined microprocessor with cache  CS/ECE 552: Introduction to Computer Architecture, Spring 2016 Used Verilog for hardware implementation to support WISC-SP13 ISA Implemented pipelining, forwarding, data and instruction set associative cache, register file bypassing, branch predictor, and carry-lookahead adders

• Line Following Robot  ECE 551: Digital System Design and Synthesis, Spring 2015 Used Verilog for FPGA programming to control a motored robot to follow a (curved) line until it reaches the specified barcode station Verilog modules includes UART, command controller, and motion controller

• Bonsai, a distributed data collection and storage system for data processing  CS 736: Advanced Operating Systems, Fall 2014 Scalable data collection in a ring structure by implementing a circular buffer for stored data with a token as synchronization logic Real-time data processing in a tree structure by multicasting map requests to backends and reducing results towards the frontend

• Parallel VLSI gate placement CAD tool on GPU  CS/ECE/EMA/ME 759: High Performance Computing for Applications in Engineering, Fall 2015 Placing connected gates closer to each other results in shorter wires, which in turn reduces delay and power consumption. Implemented gate placement algorithms for obtaining optimal wire length using CUDA and comparing it with a CPU implementation Achieved 6400 speedup with an optimized GPU implementation over CPU

• Pokemon Pallet Town Graphics  CS 559: Computer Graphics, Spring 2017 3D rendition of Pokemon FireRed Version of Pallet Town NPCs look like human figures and have hierarchical modeling movements Many textures for buildings, objects, and signs View angles and TimeOfDay affects lighting Bump mapping on grass for a more realistic 3D visualization Pokemons patrol in piecewise Bezier and Hermite curves

• Level Up Bound, a pattern-puzzle obstacle game  CS 107: Introduction to Computing and Programming, Fall 2011 Level Up Bound was an action pattern-puzzle obstacle game created by Mp(U) for Blizzard Starcraft in 2001. I remade Level Up Bound for Scratch, currently having 15 levels.