Upload
others
View
3
Download
1
Embed Size (px)
Citation preview
TITLE
Image
Accurate AMI Analysis -
Whose Responsibility Is It?
Panel Discussion: Wednesday January 20, 2015, 3:45-5pm
Moderator: Donald Telian, SiGuys
• Accurate AMI Analysis – Whose Responsibility Is It?
– Donald Telian, Signal Integrity Consultant, SiGuys
• Panel Format:
– 6 Panelists
– 5 questions
– Timed response
– The “red flag” is back
– Audience questions
Welcome to the 2016 AMI Technical Panel
PANEL:
2 IC
2 Systems
2 EDA
Accurate Analysis
EDA
IC
Systems
Whose Responsibility Is It?
your tools don’t work
go get some training
These comments
are not allowed
• Bob Miller, Avago Technologies
• Todd Westerhoff, SiSoft
• Stephen Scearce, Cisco
• Aleksey Tyshchenko, Intel
• Fangyi Rao, Keysight Technologies
• Greg Edlund, IBM
Panelists
Introduce yourself and your role at your company.
Outline your company’s and your personal
involvement with AMI models.
Question #1 of 5
BM 3:50
Todd Westerhoff
VP/Semiconductor Relations
SiSoft
• We sell Quantum Channel Designer (QCD)
• We develop IBIS-AMI models for customers
• We validate IBIS-AMI models developed by others
• We perform design work under contract using IBIS-AMI
• We actively participate in driving changes to IBIS and IBIS-AMI
Stephen Scearce Sr Signal Integrity Manager
CHG-PDS-SI Cisco Systems Inc.
• 15 years in SI/EMC and System Design at Cisco • We rely on AMI simulations for almost every product design decision
for our Serial Links • Literally hundreds of designs fabbed based on AMI Models (2.5-56G). • AMI simulations in QCD, ADS, Matlab, Hyperlynx • Personally engage with our IC vendors and Eda Vendors to work
towards amiable solutions
Introduce yourself and your role at your company.
Outline your company’s and your personal
involvement with AMI models.
Question #1 of 5
An increasing amount of design decisions are
based on AMI-based simulation results.
To what degree can we trust such analyses,
and under what conditions?
Question #2
GE bw 3:56
An increasing amount of design decisions are
based on AMI-based simulation results.
To what degree can we trust such analyses,
and under what conditions?
Question #2
AMI analysis is an abstraction of SerDes behavior, which implies limitations in representing reality
Seeming simplicity of SI simulators masks complexity of communications and probability theories
Good results are easier to like and accept compared to not so good results, but which results are true?
We can trust AMI-based analysis to the degree of:
- our understanding of the modeling approach limitations
- our understanding of the simulation tools and models
- our critical analysis of the simulation results
To what degree can we trust AMI-based analysis,
and under what conditions?
Channel model Correlation , h(t)
4100 4150 4200 4250 4300 4350
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
pulse response of TX model(eq_prec 10 eq_post 16)
AMI_INIT
AMI_GETWAVE
-0.4 -0.2 0 0.2 0.4-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
-5.5
-5
-4.5
-4
-3.5
-3
-2.5
Lab Measurement Correlation
AMI Model Correlation
Trust AMI Simulation Results? (Scearce)
Can We Trust IBIS-AMI Results?
• How good is your input data?
– Passive (channel) models?
– IBIS-AMI models?
– Reference (measured) results?
• Are simulations set up properly?
• How should simulation results be interpreted?
• We regularly see excellent correlation to measurement using a structured analysis methodology.
An increasing amount of design decisions are
based on AMI-based simulation results.
To what degree can we trust such analyses,
and under what conditions?
Question #2
Is the matrix of AMI model and analysis types manageable
or is it too large?
What types of analyses should we focus on in 2016,
and how do we get there?
Can the IBIS Committee help?
Question #3
TW 4:02
AMI Issues
• Is the matrix of AMI models and analysis types too big?
– No.
From “Understanding IBIS-AMI simulations”, DesignCon 2015
AMI Issues (Cont.)
• What should we be focused on?
– Models that accurately reflect hardware behavior and methods to document how well the two match.
• Does that mean time-domain first and statistical only if you can?
– Yes - quality comes first. There’s no point in getting a bad answer faster.
• Can the IBIS committee help?
– They can move faster and be more practical.
AMI Matrix/Issues (Scearce)
• AMI model analysis types are manageable
• Focus on the AMI models correctly depicting the analog and algorithmic behavior of the IC vs fastest solutions
• Model developers should work to model the RX algorithmic code defined in the actual silicon.
• We need RX adaptation information (or ?) in addition to height/width to verify if RX auto-adapt solution is correct.
Our team provides AMI models that support both statistical and time domain simulations
Our models allow the systems designers to choose various simulation flows
We strive to build models with good correlation between statistical and time domain simulations
This approach allows to get useful insight into system behavior with small simulation time
Is the matrix of AMI model and analysis types
manageable or is it too large?
Is the matrix of AMI model and analysis types manageable
or is it too large?
What types of analyses should we focus on in 2016,
and how do we get there?
Can the IBIS Committee help?
Question #3
What AMI improvements are you currently working on?
What AMI improvements do you need others to provide,
and specifically from whom?
Question #4
SS 4:08
2 minutes
Current Efforts & Needs (sscearce)
Cisco Efforts
• Lab BER results VS Simulation Results for specific HW/Channels
• TREND correlation for TX/RX (Brian Baek, Cisco)
• Improving accuracy of system level channel modeling/extractions
Needs
• We need AMI models to run in all EDA tools, several companies recommend only leveraging ABC as the simulation engine.
• Model suppliers to engage with EDA companies to verify models before model
delivery, Maybe generic IBIS standard for correlation (Brian B)
Integrating AMI-based simulations into the R&D flow of our SerDes IP
Making AMI models available early in the SerDes development process
Reducing models refresh time and effort based on Spice simulations
What AMI improvements are you currently working on?
What AMI improvements do you need others to provide,
and specifically from whom?
Systems designers: your constructive criticism is the strongest driver of the AMI modeling evolution
EDA vendors: sharing best known methods and practices with customers would benefit all parties
Worst-case model 225 mV x 75 ps
Worst-case lane 375 mV x 90 ps
Model-to-Hardware Correlation Case 1
Best-case model 255 mV x 80 ps
Best-case lane 810 mV x 95 ps
Model-to-Hardware Correlation Case 2
Model 200 mV x 75 ps
Hardware 190 mV x 90 ps
Lane CTLE mV ps 1 2 50 80 3 3 100 90 2 4 125 90 0 5 150 90
Measured amplitude tracks with CTLE.
Lane-to-lane loss varies ~2 dB.
What AMI improvements are you currently working on?
What AMI improvements do you need others to provide,
and specifically from whom?
Question #4
Current Efforts & Needs
Efforts
• Automated TX/RX co-optimization
• Streamlining IBIS-AMI model development
• Improving system-level post-route analysis
• Model correlation: PAM4/56Gbs and beyond
Needs
• More collaboration between customers & suppliers
• A more practical approach to IBIS enhancements
What should engineers from each discipline
(IC, Systems, EDA)
be doing to improve the quality of AMI models and analysis?
What processes and automation exist to validate AMI
models and analysis, what is lacking,
and what can we expect to see in the coming year?
Question #5
FR 4:20
3 minutes
The Curmudgeon’s View:
The Lost Art of Craftsmanship
“Craftsmanship is the quality that comes from creating with passion, care and attention to detail.” - Unknown, probably rolling over in the grave.
Characteristics of a Craftsperson
• Takes responsibility for quality of final product. • Motivated by mastery rather than social status. • Deep understanding of materials, tools and processes. • Adapts based on what is available. • Balances tradition with drive to innovate. • Plans and thinks things through first. • Relentless commitment to ongoing improvement.
Be a Simulation Craftsperson!
• Validate your data before use. – If you don’t know how – ask, experiment, learn.
• Take the time to understand your tools and processes. – Know what results you expect. Question what doesn’t look right.
• Collaborate, collaborate, collaborate. – Complex, inter-related projects and blind assumptions are not compatible.
• Quality is everyone’s responsibility!
Where do we spend our time?
0 10 20 30 40 50 60 70 80 90 100
2013-2014
2015
Goal
% of time spent in AMI/Channel/Lab
Engineering Time Engineering Time
Improve the quality of AMI models (scearce)
• IC: Focus on high quality models that accurately describe the analog, digital, and algorithmic sections of the silicon. Create test cases internally that stress the models and verify operation with all major EDA companies. Simplify/ document mapping for AMI parameters to datasheet register sets (or return register values :)
• EDA: Strive for a uniform interpretation of the AMI standard. Work to implement features that will work in all EDA tools. Tool development decisions should be based of a consensus and guide from the standard committee. Provide verification kit for each simulator to ensure operation.
• System: We need to "qualify" the model with our EDA tools to make sure it is behaving like the model creators expected. We should continue to engage with IC vendors to share correlation results to improve models across industry. Assume a lead role to provide correlation guidance to silicon companies on model requirements.
IC Vendors
- provide high-quality models correlated to Spice or Si as appropriate in the IC design phase
EDA Vendors
- provide high-quality SI simulation environments
- share best known methods and practices of SI simulator usage
Systems Designers
- understand the implications of AMI-based simulation flow
- understand the AMI models and the SI simulator of choice
- critically assess the simulation results
What should engineers from each discipline be doing to
improve the quality of AMI models and analysis?
Who’s Responsibility?
Is it…
Who’s Responsibility?
TX RX
Is it IC vendor?
Who’s Responsibility?
TX RX Channel
Is it IC vendor OR component vendor?
Who’s Responsibility?
SI Simulator
TX RX Channel
Is it IC vendor OR component vendor OR EDA vendor?
Is it IC vendor OR component vendor OR EDA vendor OR systems designer?
Who’s Responsibility?
SI Simulator
Simulation Setup
TX RX Channel
It is IC vendor OR component vendor OR EDA vendor OR systems designer!
Who’s Responsibility?
SI Simulator
Simulation Setup
TX RX Channel
AND AND AND
It is IC vendor OR component vendor OR EDA vendor OR systems designer!
Accurate AMI analysis is a shared responsibility
Who’s Responsibility?
SI Simulator
Simulation Setup
TX RX Channel
AND AND AND
Accurate Analysis
MINE!
Whose Responsibility Is It?
DT 4:40
AUDIENCE QUESTIONS
DT 4:41
TITLE
Image THANK YOU