CS250 VLSI Systems Design Lecture 7 Project Details John Wawrzynek Krste Asanovic with John Lazzaro and Brian Zimmer Lecture 7 Project CS250 UC Berkeley Fall 2011 Engineering Challenge Application Gap usually too large to bridge in one step but there are exceptions Physics Lecture 7 Project 2 CS250 UC Berkeley Fall 2011 Magnetic Compass Application Physics Lecture 7 Project 3 CS250 UC Berkeley Fall 2011 Design Abstraction Stack Application Unit Transaction Level UTL Register Transfer Level RTL Gates Circuits Devices Transistors Physics n oxi p n Conduction Band Eg Valence Band Lecture 7 Project 4 CS250 UC Berkeley Fall 2011 Properties of a Useful Abstraction Hides less important details e g for RTL don t worry how combinational logic is decomposed into logic gates Allows control of more important details e g RTL designer still controls how much logic is performed bet ween any t wo registers If done right provides portable efficiency i e same RTL can be implemented as custom logic standard cells FPGA or even vacuum tube logic with reasonably good results Lecture 7 Project 5 CS250 UC Berkeley Fall 2011 Design Abstraction Stack For complex applications that talk to internet Application and provide user interfaces these steps Unit Transaction Level UTL are way too large Register Transfer Level RTL Gates Circuits Devices Transistors Physics n oxi p n Conduction Band Eg Valence Band Lecture 7 Project 6 CS250 UC Berkeley Fall 2011 Soft ware Centric MP SoCs Almost all devices based on large ASICs need to run sophisticated soft ware MP SoC Multiprocessor System on a Chip already standard in many devices smartphones music players set top boxes games consoles digital cameras internet routers cars Typical ASIC team ratios 1 Hardware designer per 2 verification engineers per 6 soft ware engineers Lecture 7 Project 7 CS250 UC Berkeley Fall 2011 Manycore a new abstraction layer Parallel Application Program Operating System Manycore System UTL Register Transfer Level RTL Gates Circuits Devices Transistors Physics Lecture 7 Project 8 CS250 UC Berkeley Fall 2011 Manycore Abstraction Hides less important details e g programmer doesn t worry how code is compiled and executed on each processor pipeline Allows control of more important details e g parallel programmer controls how application code and data is distributed among cores If done right provides portable efficiency i e same parallel program can be executed on different multiprocessor platforms general purpose x86 platform MP SoC FPGA soft cores Lecture 7 Project 9 CS250 UC Berkeley Fall 2011 Project Topics This year s theme If the processor is the new transistor what does the standard processor library look like Assume any MP SoC will be built from a heterogeneous mix of processor types Your task is to explore variations of RISC V microprocessors in class projects i e pick a general class of RISC V processor then explore design space for that class of processors Output of class could be used to begin populating a standard processor library Rowen Tensilica Lecture 7 Project 10 CS250 UC Berkeley Fall 2011 Results from MIT Lab Exercise 6 884 2005 Pareto Optimal Points Lecture 7 Project 11 CS250 UC Berkeley Fall 2011 Quality of Results QoR QoR is term used to describe overall goodness of design Usually a multi dimensional vector including multiple metrics Area mm2 Performance Operations Second Energy efficiency Operations Joule System Level Cost Correctness not binary in practice Reliability Undetected errors Year MTBF Manufacturability Yield We will focus on first three in project Require projects to be 100 correct Lecture 7 Project 12 CS250 UC Berkeley Fall 2011 Pareto Optimal in 3D Time Operation Energy Operation Lecture 7 Project Area 13 CS250 UC Berkeley Fall 2011 Our Project Expectations B grade Single working design A grade Thorough design space exploration Lecture 7 Project 14 CS250 UC Berkeley Fall 2011 General Project Info A large scale VLSI design experience Most significant component of class 70 of grade Work in teams of 2 In exceptional circumstances might allow other team sizes 1 or 3 but have to clear with us ASAP Lecture 7 Project 15 CS250 UC Berkeley Fall 2011 Initial Proposal Due October 10 before class email to instructors TA as PDF file Not any other file format must be PDF Must include Title Team members names 2 page description of what you want to do What does the design space look like and how you will explore it Must be in PDF in case you weren t paying attention Lecture 7 Project 16 CS250 UC Berkeley Fall 2011 Project Meetings Public presentation days Each group gives 15 minute presentation to whole class Everyone must attend give feedback Private project meetings Approx 20 minutes scheduled with each group alone with staff to give private guidance Class splits into t wo sets of project groups one meets on Mondays other on Wednesdays Plus come to office hours arrange other meetings Lecture 7 Project 17 CS250 UC Berkeley Fall 2011 Public Presentations Oct 12 Present project proposal 10 mins Describe your project idea get feedback Oct 31 Nov 2 Present initial results on at least one design point 15 minute presentation max to whole class 10 slides Prepare concise informative presentation Ask class for feedback 5 minutes Nov 14 16 Present some design space exploration results Same 15 minute 5 minute discussion format Dec 5 7 Final project presentations 20 minutes each time limit strictly enforced practice Lecture 7 Project 18 CS250 UC Berkeley Fall 2011 Private Meetings Each group has a scheduled 20 minute meeting with staff to provide guidance feedback Same time slot during class time every meeting week and also used for final presentation Schedule picked randomly Cannot change unless you can convince another group to switch Come prepared with results questions Lecture 7 Project 19 CS250 UC Berkeley Fall 2011 Final Project Report Report should read like a conference paper 12 pages 10pt font one or t wo column must be PDF Report due 6AM Monday Dec 12 No extensions Email PDF file to staff Lecture 7 Project 20 CS250 UC Berkeley Fall 2011 Project Source Code Must be kept in repository so we can build test and inspect design Must be checked in before each meeting You ll want to use repo for group communication archiving Lecture 7 Project 21 CS250 UC Berkeley Fall 2011 Possible Project Ideas General Purpose processor improvements Domain specific accelerators Lecture 7 Project 22 CS250 UC Berkeley Fall 2011 layout level Standard