EEDG/CE 6301: Advanced Digital Logic
Mehrdad Nourani
Dept. of EE Univ. of Texas at Dallas
Session 01
Introduction
Motivational Discussion
Personal Products
PDA 2G/2.5G Cellular
PDAs
3D Cellular Phones
Video Phone
IP Phone
Cable Cable Modem Modem
Modem
DSL DSL Modem
Bluetooth Products
Digital Still Digital Camera Camcorder
PDA Camera
Network Still Camera
Home Networking
DAB Radio
Digital TV
Internet Audio Player
Digital Video Recorder/Server
iSTB
Technology in the Internet Age
Internet DSP & Analog
1 per Person MANY per Person!
PC Microprocessor
Minicomputer 1 per TTL/Logic Department
Mainframe Transistors 1960s 1970s 1 per Company 1980s 1990s 2000s 2010s
6
Source: UTD Seminar by Gene Frantz
PC-Age vs. Internet Age
PC Age
Computer Focus
Non-Real Time Stationary Digital I/O P and Memory
Internet Age
Communication Focus
Source: UTD Seminar by Gene Frantz
Real Time Mobile Analog I/O DSP and Analog
System-on-Chip Integration
Many Devices Per Person Lowest Cost
Analog/RF Input/Output
Internet Interface
DSP + Analog
SOC Integration
Packet
Processor
SOC integration means more than integration of multiple IP digital cores. It means integration of functions that are implemented today in different technologies.
Source: UTD Seminar by Gene Frantz
Technology in the Internet Age
Moores Law is predicted to stagnate toward the end of the decade but SOC Integration has the potential to continue IC cost reduction and to perpetuate growth of Personal Internet Products.
Source: UTD Seminar by Gene Frantz
11
Amazing Growth
1971: Intel 4004
2300 Transistors, 750KHz, 60000 Operations/sec $200
2000: DEC Alpha
15.2 Million Transistors, 700MHz,2 Billion Operations/sec $300
Gordon Moore (Intel founder):
If the car industry had been progressing at the rate the semiconductor industry has, today a Rolls-Royce would cost three dollars, could get half-a-million miles per gallon of gas, and would be cheaper to throw away than to pay for parking.
12
What is Our Course About?
We need to have a good understanding of three concepts:
VLSI
Technology
Design
Technique
System
Methodology
13
SoC: System on a Chip
Prediction: Many SoCs will have > 1 Billion gates How do you create million gate ASICs with same amount of resources? While
Decrease development time Increase functionality and performance Keep small design teams
Important Issues
Design Methodology (Design flow) Tools that support the Methodology IP reuse (Intellectual Property)
14
Modeling Types Behavioral model
Explicit definition of mathematical relationship between input and output No implementation information It can exist at multiple levels of abstraction
Dataflow, procedural, state machines,
Structural model
A representation of a system in terms of interconnections (netlist) of a set of defined component Components can be described structurally or behaviorally
15
The Y-Chart - Levels of Abstraction
16
The Y-Chart - Terminology
17
Example: Various Models of NAND Gate
O <= NOT ( A1 AND B1);
Behavioral Model: (Boolean Equation)
Structural Model: (Logic Gate)
Physical Model: (Transistors or Layout Mask)
19
Importance of VLSI Test
21
Importance of VLSI Testing Test cost will be dominant in this decade [ITRS01]
Cost: Cents/Transistor
New DFTs
ITRS
23
Testing for Signal Integrity
Process Variations Shrink of Technology Increase of Frequency Physical Defects
Smaller Design Rules
Wave-Oriented Phenomena
There are only two kinds of designers: the ones who have signal integrity problems, and the ones who will.
[www.chipcenter.com]
25
Fatal Problems on First Spin
Overall 61% of new ICs require at least one re-spin
45 40 35 30 25 20 15 10 5 0 Functional Signal Error Integrity Reliability High Power
Analog Tuning Clock Skew Mixed Signal IR Drop Others
[www.deepchip.com]
Firmware Error
26
Importance of CAD
30
Role of CAD Tools
CAD tools improve productivity Definition: A software program which assists in performing or automates a particular design function Major types: Schematic capture Analyzers/Simulators Synthesis
31
Design Complexity and Productivity
Moore's Law
Design Productivity Time (Months)
32
CAD Design Approaches
Goal of each CAD design flow methodology is to increase productivity of the design engineer Increasing the abstraction level of the design methodology and tools is one approach:
Gates/eng./month 1.5K - 6K Describe-synthesize
Abstraction Design Data
Design Sizes > 1M gates
300 - 600
Schematic Capture-simulate
100K - 500K gates
34
Schematic Capture
wire
labels
Gate primitive
35
Schematic Capture Editor Definition: An editor which can be used to create and display an interconnected set of graphic tokens. Graphic token types:
primitives (built-in) new models
Uses of schematics
simulation wiring
36
Simulators
Definition: A program which models the response of a system to input stimuli. Types: deterministic and stochastic. Simulation is used to establish design correctness (70% of design time). A model underlies simulation.
37
Factors Affecting Design Efficiency
Programming Method Computer Architecture Level of Abstraction Rule: Simulate at the highest level you can, and still get desired information. Rule-of-10: It will be 10 times more costly if the problem is discovered in the next level.
38
Importance of HDL
48
Effect of Hardware Description Language (HDL)
The Complexity and Size of Digital Systems leads to Breadboards and prototypes which are too costly Software and hardware interactions which are difficult to analyze without prototypes or simulations Difficulty in communicating accurate design information Want to be able to target design to a new technology while using same descriptions or reuse parts of design (IP)
49
Using HDL in Design
Source: VHDL (Z. Navabi - McGraw Hill)
50
Using HDL in Simulation
52
Using HDL in Synthesis Tools
53
Using HDL in Synthesis & Optimization
54
The Challenge of Optimization
55
Coping with NP-hard Problems
In system level design we confront many NP-hard optimization problems. Simpler sub-problem based on dominate cost or special problem structure problems exhibit structure
optimal solutions found in reasonable time in practice
approximation algorithms heuristic solutions high density of good/reasonable solutions?
56
Not a Solved Problem
NP-hard problems
almost always solved in suboptimal manner or for particular special cases
decomposed in suboptimal ways quality of solution changes as dominant costs change (relative costs are changing!) new effects and mapping problems crop up with new architectures, substrates
57
Decomposition/Partitioning
Easier to solve
only worry about one problem at a time
Less computational work
smaller problem size
Abstraction hides important objectives
solving 2 problems optimally in sequence often not give optimal result of simultaneous solution Question: Like what?
58
Partitioning
Definition: Given a set of objects O={o1,,on} determine a partition P={p1,,pm} such that p1UUpm=O, pipj= for all i,j, i#j and the cost determined by an objective function f(P) is minimal. NP-complete for general graphs/problems Many heuristics/attacks System designer must do two things:
1. Selecting a set of system components (allocation) 2. Partitioning the systems functionality among those components (partitioning).
Partitioning Issues:
Abstraction level Granularity Estimation
62
Partitioning Heuristic
Greedy, iterative
pick one partition that decreases cost (i.e. a user defined metric) and move it repeat
Small amount of:
look past moves that make locally worse randomization
Estimation Metrics:
Fast (usually analytical) estimate of area,time,power,etc. Fidelity of estimation
Quality Metrics:
Hardware/software cost, performance, benchmarking
63
Design Space
64
Concept of Design Space
There exists no perfect/optimal algorithm for the design of complicated systems The designer moves around in a space The coordinates of the space are optimization criterion: speed, chip area, cost, power, pins, etc. Motion in the space involves tradeoffs
65
A 3-Dimensional Design Space
a design
66
Example: Speed-Area Tradeoff
67
Example: Workstation Cost/Speed Tradeoff
Cost ($)
(C2,S2)
C1 $ 5K
S1 C2
50 MIPS $ 30K
500 MIPS $ 10K 280 MIPS
(C3,S3) (C1,S1)
S2 C3 S3
Speed (MIPS)
68
