Академический Документы
Профессиональный Документы
Культура Документы
Lecture 1
Lecture 1
Design and Technology Trends
Overview
Lecture 1
Recent Trends
1.5GHz Itanium chip (Intel), 410M tx, 374mm2 , 130W@1.3V
1.1 GHz POWER4 (IBM), 170M tx, 115W@1.5V
if these trends continue, power will become unmanageable
Lecture 1
Multi-Core Design
Intels 80-core chip
In 65-nm technology with 80 single-precision, floating point
cores delivers performance in excess of a teraflops while
consuming less than 100 w.
A 2D on-die mesh interconnection network operating at 5
GHz provides the high-performance communication fabric to
connect the cores.
Interconnects are the biggest bottleneck
We need to look beyond the metal/dielectric-based planar
architectures
Optical, 3D integration and Wireless are the emerging
alternatives
Lecture 1
Multi-core applications
Nokia Sparrow
Intel LARRABEE
Lecture 1
2D IC
3D IC
6
Lecture 1
6
Photonic Communication
High bandwidth photonic links for high payload transfers
Limitations on switch architecture
More than 4-port designs are complex
Lecture 1
Lecture 1
Optical Interconnects
Lower Latency
and Energy
Dissipation
Lecture 1
Technology Node
set by 1/2 pitch
(interconnect)
Poly width
Gate length
(transistor)
Lecture 1
10
Before Scaling
After Scaling
Channel Length
L = L * s
Channel Width
W = W * s
tox = tox * s
Junction depth
xj
xj = xj * s
Power Supply
Vdd
Vdd = Vdd * s
Threshold Voltage
Vth
Vth = Vth * s
Doping Density, p
n+
NA
ND
NA = NA / s
ND = ND / s
Lecture 1
11
1999
2001
0.7x
2004
2007
2010
2013
2016
2019
0.7x
N+1
Two year cycle between nodes until 2001, then 3 year cycle begins.
Lecture 1
12
1000
100
80386
80486
Pentium
PentiumII
10
Dec-83
Expon.
Dec-86
Dec-89
Lecture 1
Dec-92
Dec-95
Dec-98
13
10000
Forward projection
may be too optimistic
P4
1000
100
80386
80486
Pentium
Pentium II
10
Dec-83
Expon.
Dec-86
Intel: Borkar/Parkhurst
Dec-89
Dec-92
Dec-95
Lecture 1
Dec-98
Dec-99
Dec-00
Dec-01 Dec-02
14
100.00
80386
80486
Pentium
PentiumII
Expon.
10.00
Dec-83
Dec-86
Dec-89
Lecture 1
Dec-92
Dec-95
Dec-98
15
where is ratio of
5
=0.0
Delay
=0.5
=1.0
=2.0
Parasitic output
Capacitance to gate
capacitance
0
0
Fanout
1X
4X
16X
CIN
Cload
Lecture 1
1000ps = 25 FO4
Data
L o g ic
Q
N
C lk
40ps
C lk
Clock
(b)
(a)
Lecture 1
17
Transistors Double
Every Two Years
10,000
1,000
100
100
Transistors
(MT)
10
P6
486
Pentium proc
386
0.1
286
8085
0.01
0.001
Source: Intel
70
4004
2X Growth
in 2 Years!
8086
8080
8008
80
90
Lecture 1
00
10
18
100
28
32
36
41
486
10
Pentium proc
386
8080
8008
4004
8086
8085
286
1
70
80
90
00
10
Source: Intel
Lecture 1
19
10,000
1,000
Nuclear Reactor
HotPentium
Plate
100
processors
Power
(Watts)
286
8086
10
8085
8080
486
386
8008
4004
0.1
Source: Intel
71
74
78
Lecture 1
85
92
00
04
08
20
Speed == Power
10W/cm2 limit for convection cooling, 50W/cm2 limit for forced-air cooling
Large currents, large power surges on wakeup
Die size will not continue to increase unless more memory is used to occupy
the additional area
additional power dissipation coming from subthreshold leakage
Clocks cannot be generated with period < 6-8 FO4 INV delays
Around 14-16 FO4 INV delays is limit for clock period
21
MP3
JPEG
Simple Moving Picture
Available Time
6-10Hr
CPG
PWR
PWM RTC
CPU
FICP SSP
Sound
I2C
USB
USB OST
GPIO
MMC
KEY
UART AC97
4 48MHz
I-cache D-cache
32KB 32KB
6.5MTrs.
Max 400MHz
DMA controller
MMC
Peripheral Area
Processor
Area
I2S
MEM
Cnt.
LCD
Cnt.
32MB
Data Transfer
Area
100MHz
22
23
50
Random Logic
40
Memory
30
Analog
20
Cores
10
0
1999
2000
Lecture 1
2001
24
Year
Technology
Chip Complexity
ASIC Frequency
1997
250 nm
50M Tr.
100MHz
1999
180 nm
150M Tr.
200MHz
2002
130 nm
250M Tr.
400MHz
2004
90 nm
500M Tr.
600MHz
Lecture 1
25
Gates Required
Gates/Day (Verified)
Total Eng. Days
Total Eng. Years
Cost/Eng./Year
Total People Cost
Other costs (masks, tools, etc.)
~12.5M
1K (including memory)
12,500
35
$200K
$7M
$8M
Lecture 1
26
Productivity Gap
Lecture 1
27
Lecture 1
28
MEM
Hardware IP
SW IP
Application
Space CPU
FPGA
Processor(s), RTOS(es)
and SW architecture
Methodology / Flows:
Programmable IP
*IPcan
canbe
behardware
hardware(digital
(digital
*IP
oranalog)
analog)or
orsoftware.
software.
or
IPcan
canbe
behard,
hard,soft
softor
or
IP
firm(HW),
(HW),source
sourceor
or
firm
object(SW)
(SW)
object
Scaleable
bus, test, power, IO,
clock, timing architectures
System-level performance
evaluation environment
HW/SW Co-synthesis
SoC IC Design Flows
Foundry-Specific
Pre-Qualification
Lecture 1
Lecture 1
30
Syllabus
Three broad categories
System on chip design and design for testability
Role of interconnects in contemporary SoC Design
Importance of Power and Low power SoC design
methodology
Lecture 1
31
References
Lecture 1
32
Assignments
There will be several homework and reading assignments. In
reading assignments students are expected to read research
papers and submit summaries. The reading list will be available
on the course website. In class, you will be told which papers
you should review.
Each student will have the opportunity to present one paper to
the class. The list of papers will be available in the course
website. Each student should choose one of the listed papers.
Lecture 1
33
Project
Lecture 1
34