332:437 Concepts in Digital Systems Design

1
332437Concepts in Digital Systems Design

• Instructor Prof. Michael L. Bushnell
• Teaching Assistants
• Ms. Srihitha Yerabaka
• Mr. Raghuveer Ausoori
• Course web site
• http//www.caip.rutgers.edu/bushnell
• ECE Department Rutgers University

2
Changes in ECE Undergrad. Hardware Courses
3
Course Structure

• Uses Verilog
• No more troubles with arithmetic conversions
• Less trouble with configurations
• Course is now a 2-term sequence
• 1st term Academic content
• 2nd term Design Project (exclusively)
• Use automatic logic synthesis to synthesize
  project as Field-Programmable Gate Array

4
332437 Lecture 1History of Fault Tolerance

• Motivation for fault tolerance
• Generations of Computers
• Fault Tolerance
• Definitions
• Applications
• Triple Modular Redundancy
• Mid-Value Select Technique Flux Summing
• Summary

Material from Design and Analysis of Digital
Fault Tolerant Systems, By Barry Johnson,
Addison-Wesley Publishers
5
Photographs from NASA

• Made possible by fault-tolerant interplanetary
  spacecraft (unmanned) and the Hubble space
  telescope
• Spacecraft Electronics
• Severe radiation fields (Gamma rays)
• Extremes of heat and cold
• Cosmic rays
• Extreme electromagnetic disturbance (particularly
  around Jupiter)

6
Hubble Space Telescope Rising from Space Shuttle
7
Neptune Rising Above Surface of its Moon
8
The Orion Nebulae
9
Cartwheel Galaxy
10
4 Largest Moons of Jupiter -- Ganymeade,
Callisto, Io, and Europa Voyager II
11
New Asteroid Found by Voyager II
12
Voyager II Photograph of Neptune
13
Voyager II -- Rings of Saturn
14
Voyager II -- Rings of Neptune
15
Device and System Reliability
16
Dependability-Performance Trade-off
17
Operational Times for Fault Tolerant Mobile
Computers

• Simplex Single processor, no fault-tolerant
  hardware
• TMR Triple Modular Redundancy (described later)
  fault-tolerant hardware
• Uses a lot of power

18
Generations of Computers

• 1st -- All electronic, stored program 1945-55
• Manchester Mark I Kilburn
• Princeton IAS von Neumann
• Univac I Mauchley Eckert
• Fault-tolerance needed for computer to work at
  all
• 2nd Discrete Transistor Computers 1955-64
• IBM 7090/7094
• Univac 1100
• MIT Whirlwind Forrester, Wang, Olsen
• Invention of Magnetic Core Memory
• Invention of CRT Display with light pen

19
Generations of Computers

• 3rd Invention of Integrated Circuit 1964-74
• 1959 Kilby, Texas Instruments
• 1960 Noyce, Fairchild
• IBM System/360 Blaauw Brooks
• Single architecture, whole range of
  price/performance
• 4th Invention of Dynamic RAM Memory 1974-88
• Invention of mprocessor (Intel 4004)
• IBM System/370
• DEC VAX 11/780
• DRAM replaces magnetic core memory
• Virtual Memory (segmentation and paging) used

20
Generations of Computers

• 5th Parallel Distributed Computers
  1988-present
• Enabled by cheap VLSI Hardware
• Pervasive Computer Networking
• Carnegie-Mellon c.mmp 16 processor DEC pdp-11
• Distributed memory, crossbar interconnect
• IBM SP-2 1 to 64 processors
• Sun Work Station
• IBM PC
• N-Cube 10 1024 processors

21
Current Ultra-Large Scale IC Technology

• AMD K8 233 million transistors
• Intel Smithfield -- 230 million transistors
• New IBM Cell Chip mprocessor 234 million
  transistors
• Hardware on chip doubles every 1 ½ years
• Currently
• 2.8 GHz mprocessor clock rates
• 1 Billion transistors on a chip
• Beginning of Network-on-a-Chip Era

22
Major Changes in Systems Design

• Fault-Tolerant Computing now affordable and
  necessary for mobile sensor-on-silicon
  applications for medicine, networking
• Severe problems in verification and testing of
  computers
• 60 of cost of hardware, Intels biggest capital
  cost
• Hardware cannot be tested unless specific design
  procedures followed
• Network reliability now a major headache
• Low-Power Design most important hardware
  problem
• System-on-a-Chip put several mprocessors (e.g.,
  8086 DSP), DRAM, glue logic, A/D, D/A, analog
  filters, chemical sensors, wireless
  transmitter/receiver on 1 chip

23
Definitions

• Fault-tolerance System can continue correct
  performance in presence of hardware/software
  faults
• Fault Physical defect or flaw in
  hardware/software component
• Error Manifestation of a fault
• Failure situation where the error resulted in a
  a system incorrectly performing its function

24
3 Universe Model (Avizienis)

• Physical Universe semiconductor, power supply,
  printer, etc.
• Informational Universe where errors occur
• Incorrect computer data words
• Incorrect digital voice/picture image
• External or Users Universe where user of
  system ultimately sees effects of faults errors
• Fault-latency time between occurrence of fault
  and appearance of an error caused by that fault
• Error-latency time between appearance of error
  and appearance of resulting failure

25
Fault Causes

• Specification mistakes wrong algorithms,
  architectures, hardware/software specifications
• Implementation mistakes poor design, poor
  component selection, poor construction, software
  coding errors
• Component defects manufacturing main cause of
  faults
• Imperfections, random defects, wear-out (broken
  bonds, corrosion)
• External disturbance g rays, a particles,
  electromagnetic interference, battle damage,
  environment extremes

26
Fault Description

• Nature hardware/software/analog/digital
• Duration how long fault is active
• Permanent in existence indefinitely
• Transient appears/disappears in very short time
• Intermittent appears/disappears/reappears
  repeatedly
• Extent localized to a given hardware or
  software module or globally affects hardware
  /software/both
• Value
• Determinant status in unchanged throughout time
• Indeterminant status at time T may differ from
  status before or after T

27
Fault Tolerance Methods

• Fault Tolerance give system the ability to keep
  performing its tasks after faults occur
• Fault Avoidance -- Prevent fault occurrence
• Design reviews, component screening, testing
• Fault Masking prevent system faults from
  introducing errors into system informational
  structure

28
Methods to Achieve Fault Tolerance

• Fault Masking
• Reconfiguration eliminate faulty module
  restore system to operation
• Fault Detection recognize that fault occurred
• Fault Location find where fault occurred
• Fault Containment isolate fault prevent from
  spreading through system
• Fault Recovery remain operational or regain
  operation after faults

29
Metrics

• Reliability R (t) conditional probability that
  system works throughout t0, t, given that it
  worked at t0
• Note 0.97 0.9999999
• Availability A (t) probability that system is
  available at time instant t perform its function
• Unreliable but available
• Safety S (t) probability that system will
  perform its function correctly or will
  discontinue working in a way that does not affect
  operation of other systems or endanger people

30
Metrics (continued)

• Performability P (L, t) probability that system
  will be at or above level L of performance at
  time t
• Graceful degradation
• Maintainability M (t) probability that a failed
  system will be restored to operation within time
  period t
• Testability ability to test for system
  attributes controllability observability
• Design for Testability Now critical to
  construct any digital system if it is to be
  successfully manufactured
• Method Add hardware strictly for testing
  purposes
• Dependability reliability availability
  performability testability

31
Fault-Tolerant Computing Applications

• Long-life applications
• P (operation at time t 10 yr.) 0.95
• Satellites Unmanned Space Flight
• ATT Telstar Communications Satellites
• NASA Martian Pathfinder
• Critical Computation
• NASA Space Shuttle
• Foxboro Nitroglycerine Plant Controls
• Maintenance Postponement
• Lucent 5 ESS Telephone Exchange
• High Availability
• New York Stock Exchange Quotron System

32
Redundancy Techniques

• Passive fault masking (hide faults)
• Active or dynamic detect fault remove broken
  hardware from system with electronic switch
• Hybrid Combine 1 2

33
Passive Redundancy

• Triple Modular Redundancy
• Single point of failure, restoring organ
• Generalize N modular redundancy
• N must be odd if majority voting is used
• Problems
• Very expensive
• 3 results may not agree e.g., in analog control
  system, A/D converters have jitter in least
  significant bits may disagree
• Solve Mid-Value select technique

34
Triple Modular Redundancy (TMR)
35
TMR with Triplicated Voters
36
Software Voting
37
Mid-value Select Technique
38
Passive Redundancy (continued)

• Frequently, one result must be produced
• Leads to single point of failure problem
• Example Motor Controller
• Solve
• Flux summing used closed loop control system to
  compensate for faults
• Secondary current a S primary currents
• Works because flux summer transformer is
  incredibly reliable

39
Flux-Summing
40
Summary

• Motivation for fault tolerance
• Generations of computers
• Fault Tolerance
• Applications
• Triple Modular Redundancy
• Mid-Value Select Technique Flux Summing
• Fault tolerance necessary for applications
• Medicine
• Transportation
• Defense
• Inter-Planetary Exploration