Breaking the Wall of Interconnect: Research and Education
description
Transcript of Breaking the Wall of Interconnect: Research and Education
![Page 1: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/1.jpg)
1
Breaking the Wall of Interconnect: Research
and Education
Chung-Kuan ChengCSE DepartmentUC San DiegoCkcheng at ucsd.edu
EDA Education and Research Workshop at ICCAD 2008
![Page 2: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/2.jpg)
2
Outlines
• Technology Trend
• Interconnect: RC Segments + Buffers
• Interconnect: Transmission Lines + Buffers (On-Chip)
• Research and Education Directions
• Hands-on Experiences
• Conclusion
![Page 3: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/3.jpg)
3
Technology Trend (ITRS 2007)
![Page 4: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/4.jpg)
4
Technology Trend
![Page 5: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/5.jpg)
5
90nm 45nm 22nm
global
wire
m1
wire
global
wire
m1
wire
global
wire
m1
wire
![Page 6: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/6.jpg)
6
Technology Trend: Wire Scaling
![Page 7: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/7.jpg)
7
Technology Trend: Packaging
![Page 8: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/8.jpg)
8Delay comparison of wires (metal 1 and global) vs. gates
Interconnect: RC Segments + Buffers
![Page 9: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/9.jpg)
9Energy comparison of wires (metal 1 and global) vs. gates
Interconnect: RC Segments (Energy per Bit)
![Page 10: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/10.jpg)
10Delay energy product comparison of wires (metal 1 and global) vs. gates
Interconnect: RC Segments (Delay Energy Product)
![Page 11: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/11.jpg)
11Bandwidth comparison of wires (metal 1 and global) vs. gates
Interconnect: RC Segments (Bandwidth)
![Page 12: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/12.jpg)
12
Repeated RC Segments• Analytical formula based flow implemented in MATLAB [3]
• Parameters: • a=0.4, b=0.7: constants related to transistor switching model• f=1: ratio of diffusion capacitance to gate capacitance of transistor• g=1.34: P/N ratio of transistor width
• Technology variables:• r0: output resistance of min-sized inverter; rw: wire resistance per unit length• cmos: min-sized NMOS gate capacitance; cw: wire capacitance per unit length
• Design objects: min-d, min-dp, min-d2p
node r0(kOhm) cmos(fF) rw(Ohm/um)* cw(pF/m)* Vdd(V)
Pitch(nm)
90nm 11.87 0.140 0.262 148.2 1.1 410
65nm 12.68 0.079 0.907 133.9 1.0 210
45nm 9.09 0.076 2.099 128.3 1.0 135
32nm 10.36 0.039 3.979 112.2 0.9 96
22nm 9.45 0.019 8.081 105.2 0.9 66 * Data are computed using formula in [3] based on wire parameters of ITRS report 2007 [1]
![Page 13: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/13.jpg)
13
Glossary
![Page 14: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/14.jpg)
14
Glossary (cont’d)
![Page 15: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/15.jpg)
15
Performance metrics of repeated wire
• Normalized delay:
• Normalized power (energy/bit):
– Leakage power factor:
• Normalized delay power product:• Bandwidth:
0 0(1 )(1 )(1 )stage nmos w
n w w inv w nmos invinv inv inv
d b g f r c br cdelay ar c l b g r c s
l l s
2(1.1 )(1 )(1 )stage leak nmos invn w dd
inv inv
p f g c spower c v
l l
![Page 16: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/16.jpg)
16
Optimum repeater insertion for minimizing delay
• Optimum repeater interval and size:
• Optimum delay and power
![Page 17: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/17.jpg)
17
Interconnection: Transmission Lines
Scheme 1: Repeated RC wire
Scheme 2: On-chip T-line driven by inverter chain
Scheme 3: On-chip T-line driven by inverter chain w/ resistive termination
2RL
2RL
Distributive RC wire
Transmission Line
Transmission Line
![Page 18: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/18.jpg)
1818
Theory (Telegrapher’s Equation)
• Telegrapher’s equation:
),(),(),(
),(),(
),(
tzGVdt
tzdVC
dz
tzdIdt
tzdILtzRI
dz
tzdV
• Propagation Constant:
jCjGLjR ))((
• Wave Propagation: zjzeVzV 0)(
• Alpha and Beta corresponds to speed and phase velocity.
![Page 19: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/19.jpg)
19
• Operation Region– RC region:
– LC region:
• Two parameters used to verify the region[17] – Upper bound of wire length for lumped element modeling
– Corner frequency between RC and LC region
On-Chip Transmission Line (1)
19
![Page 20: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/20.jpg)
20
T-line Structure and Extraction
tan 0.00068
W
H
T
H
S
G
G
S2G S1
3.1r 62.2 10Cu cm
S3 G
• Single-ended strip-line configuration– Wire length: 5mm
• Extraction includes 3 adjacent wires to consider crosstalk– Use H=1.2um(2.4um) for C extraction and H=4.4um(8.8um) for L
extraction, to consider the worst case
![Page 21: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/21.jpg)
21
Experimental Settings
• Tools– 2D EM-field solver CZ2D from EIP tool suite of IBM
• Frequency dependent RLGC tabular model– HSPICE with predictive transistor model
• Synopsys level3 MOSFET model– Design flow is implemented in MATLAB
• Case configuration– Study and compare the performance metrics of 3 schemes at 45nm
node– Using the worst-case input pattern -+- to simulate delay/power. – Optimize under 3 object functions: min-d/min-dp/min-d2p
![Page 22: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/22.jpg)
22
Wire Bandwidth and Eye-Opening
Effect of driver impedance on wire bandwidth
Effect of driver impedance and termination resistance on wire output eye-opening
![Page 23: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/23.jpg)
23
Performance Comparison• Rs:10ohm, Wire:16X for scheme 2 and 3
![Page 24: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/24.jpg)
24
Effect of RS on Step Response
Wire 16X, w/ termination resistance Rload=220ohm
![Page 25: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/25.jpg)
25
Effect of Rload on Step Response
Wire 16X, w/ termination resistance RS=10ohm
(Optimal Value)
![Page 26: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/26.jpg)
26
Eye-diagrams of T-line scheme
• w/ termination resistance
• Rload=220ohm
• Optimal solution of min-ddp
![Page 27: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/27.jpg)
27
Interconnect Dominated Designs: Research Directions
• Analysis: Simulation from wires, circuits, to systems– Wires and device elements: modeling, extraction and measurement – Circuits:
• Spice simulation of whole circuits• Power analysis• Timing analysis (LOCV)
– Systems: • Buses and interfaces• Function and logic analysis• Rapid prototyping and emulation• Multi-domain analysis: EM, thermal, mechanical, biological analysis
• Synthesis: Design from wires, buses, layouts, modules, to networks– RC segments, transmission lines, photonic communication– Power systems, clock distributions, signal buses– Physical layout: floorplanning, placement and routing– Function module synthesis– Network architectures
![Page 28: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/28.jpg)
28
Interconnect Dominated Designs: Education
• Computer Sciences– Software Engineering– Algorithms and Numerical Methods– Logic and Arithmetic Designs– Computer and Network Architectures– Distributed Computation
• Electrical Engineering– Physics (Photonics)– EM Waves– Circuit Theory– VLSI Designs
• Motivations and Methodologies
![Page 29: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/29.jpg)
29
Education: Motivations and Methodologies
• Data Mining: Literature, Patents, Products, Packages, Research Groups
• Problem Solving – Statement of the Problems– Hands-on Experience– Debugging
• Communication– Teaming– Networking– Broadcasting
![Page 30: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/30.jpg)
30
Education: Hands-on Experience
• Y. Zhu, T. Weng, C.K. Cheng, "Enhancing Learning Effectivelness in Digital Design Courses by Programmable Logic Boards," to appear in IEEE Trans. on Education.
• T. Weng, Y. Zhu, and C.K. Cheng, "Digital Design and Programmable Logic Boards: Do Students Actually Learn More?" ASEE/IEEE Frontiers in Education Conf., Session S1H-1, pp. 1-6, 2008.
![Page 31: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/31.jpg)
31
Education: Hands-on Experience• Introduction
• Class Information
• Teaching Experience
• Students Feedback
• Final Exam Results
• Conclusion
![Page 32: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/32.jpg)
32
Hands-on Experience: Example• Digital design is an essential part of the
CS curriculum• Challenges in teaching
– Lack of previous hardware class work– Lack of interest in hardware among CS
students
• EDA Software Process vs Hardware Execution
![Page 33: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/33.jpg)
33
Hands-on Experience: Example• Concerns
– Whether CS students have proper background?– How difficult to teach VHDL and integrate it into
courses?– Board prices & Teaching load
• Benefits– Students can gain better understanding, grasp full
design implementation cycle (Zema 1998, Areibi 2001, Vera et al. 2006, etc.)
![Page 34: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/34.jpg)
34
Class Structure• Introductory digital logic design course in
CS program at UCSD– Combinational & sequential logic– Standard modules: decoders, MUX…– System design
• Associated labs– Altera Quartus II software– UP-2 CPLD board (100 US$)– Schematic & VHDL designs
![Page 35: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/35.jpg)
35
Lab Assignments• 1. Combinatorial Circuit Design:
5 basic circuits (adder, multiplexer)
• 2. Sequential Circuit Design: Shift registers, counters, clock
• 3. Finite State Machines:traffic light, train controller, grey counter
• 4. CPU Design
![Page 36: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/36.jpg)
36
Teaching Goals & Methodologies
• 1. Utilize the PLD board to promote learning for the students– Design labs to reflect the teaching materials
• 2. Design labs that are enjoyable and educational for the students– Design labs that are novel and have a
degree of fun to them
![Page 37: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/37.jpg)
37
The PLD Board
![Page 38: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/38.jpg)
38
Students Feedback• Survey structure
– 16 statements to evaluate (1: strongly disagree, 5: strongly agree)
– 14 short-answer questions
• Conducted in the end of the term
• 35 returned out of 38 students (92.1%)
![Page 39: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/39.jpg)
39
Students Feedback 1 – PLD board usefulness
• 86% enjoyed using the PLD board
• 9% enjoyed running simulations
• 80% felt that they learned digital logic from using the board
• 23% felt they learned from using the simulations
• 77% agree PLD board helped them learn digital design
![Page 40: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/40.jpg)
40
Students Feedback 1 – PLD board usefulness
• 63% stated that they would not have learned the contents as well without the board
• 26% said they would have spent a lot of time and effort to grasp the knowledge without the board
• 4% thought the boards were not useful
![Page 41: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/41.jpg)
41
Students’ Words• “I thought it was really neat to be able to see
the number 32 instead of 10000 with high and lows in a timing diagram”
• “it (the PLD board) allowed me to apply concepts learned in the class in a real world situation and understand how people in the industry go about using these concepts to solve these problems”
• “(The CPU design project is enjoyable) because I got to build an actual CPU so that I could get a feel of what low level programming is like”
![Page 42: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/42.jpg)
42
Students’ Words (cont.)• “This class is very useful and I have learned a
lot from it, especially lab 4 because I really got to see how things are done”
• “At first I thought it would just be another class to drag myself to everyday. However, I now have a great appreciation for digital design. So much so that I am considering a career on it.”
![Page 43: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/43.jpg)
43
Final Exam Score Comparison• Compare the final exam results on
CSE140 (Digital Design Theory) between Fall 2005 and Spring 2007
• Similar course structures, similar final exams, same instructor
![Page 44: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/44.jpg)
44
Final Exam Score Comparison
![Page 45: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/45.jpg)
45
Observations• The second tier students benefit most• Survey comments indicate that the use of board
solidified many of the concepts• From students who scored 80-90
– 100% agreed that they learned more from the board– All thought tutorials are useful– All but one student said the course was more
interesting than they originally expected
![Page 46: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/46.jpg)
46
Observations• We investigated whether using a programmable
logic board helps CS students• We introduce PLD boards in our labs• Students answered a detailed survey
– They enjoyed the boards– Those struggled in the class benefit most– Tutorials are important
• Exam results comparison indicates that middle score range students improved a lot from the boards
![Page 47: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/47.jpg)
47
• Technology Trend: Interconnect Dominance
• Expansion of Literatures and Tools
• Classic Theories and New Problems Covering Multiple Disciplines
• Motivations and Methodologies
• Hands on Experiences
Summary
47
![Page 48: Breaking the Wall of Interconnect: Research and Education](https://reader036.fdocuments.us/reader036/viewer/2022081514/56814c26550346895db92cf8/html5/thumbnails/48.jpg)
48
Thank you!Q & A