competitive solutions
DESCRIPTION
Competitive Solutions. DX0120. Nature of Assignment. Retained by FTC counsel Spring 2002 Conduct economic analysis of Rambus’s alleged actions Analyze competitive nature of Rambus’s alleged conduct Determine competitive effects or potential effects of such conduct - PowerPoint PPT PresentationTRANSCRIPT
Competitive SolutionsCompetitive Solutions
DX0120
Nature of AssignmentNature of Assignment
Retained by FTC counsel Spring 2002
Conduct economic analysis of Rambus’s alleged actions
Analyze competitive nature of Rambus’s alleged conduct
Determine competitive effects or potential effects of such conduct
Determine appropriate remedies
DX0121
Understanding of FTC AllegationsUnderstanding of FTC Allegations
Subversion of open standard process (JEDEC) through non-disclosure of IP and related misrepresentations
Industry adoption of JEDEC standards embodying claimed Rambus IP
Subsequent enforcement of Rambus IP against JEDEC-compliant DRAM
Rambus’s conduct eliminated alternative technologies that, before the standards were issued, were commercially viable
Harm to competition in / monopolization of DRAM technology markets
DX0122
Key Economic QuestionsKey Economic Questions
1. What are the relevant antitrust markets?
2. Does Rambus possess substantial market / monopoly power in such markets?
3. Did Rambus acquire market / monopoly power through exclusionary conduct?
4. Has Rambus’s conduct resulted in anticompetitive harm (actual or threatened)?
5. What remedy (if any) is needed to restore competition / alleviate the anticompetitive effects of Rambus’s conduct?
DX0123
Interviews ConductedInterviews Conducted
DRAM engineers
DRAM plant managers
JEDEC participants
DRAM users
DX0124
Materials Reviewed / Relied Upon Materials Reviewed / Relied Upon
Business records from Rambus and third parties
Minutes and presentation materials from JEDEC
meetings
Relevant deposition / trial testimony
Reports of FTC / Rambus experts
Relevant economics literature
Literature on DRAM and semiconductor industries
Publicly available trade press, analyst reports, and
company sources
DX0125
Case StudyCase Study
Methodology: Comprehensive review of public /
private reports and information Focus: evolution of DRAM standards / technologies
Time period: 1990 to present
Information sources: publicly available materials,
trade press, analyst reports, discovery materials
Purpose: assessing economic factors influencing
choices among alternative DRAM technologies /
standards
DX0126
Summary of ConclusionsSummary of Conclusions
1. What are the relevant antitrust markets?
Conclusions:
Four relevant technology markets
Each consists of commercially viable alternatives for addressing specific DRAM design issues
Each market is world-wide in scope
The four markets can be viewed together as a single “cluster market”
DX0127
Summary of ConclusionsSummary of Conclusions
2. Does Rambus possess substantial market / monopoly power in such markets?
Conclusions:
Rambus possesses monopoly power in each of the relevant technology markets
Rambus’s monopoly power stems from the use of its patented technologies in the dominant world-wide DRAM technology standards (i.e., JEDEC’s SDRAM and DDR-SDRAM standards)
DX0128
Summary of ConclusionsSummary of Conclusions
3. Did Rambus acquire market / monopoly power through exclusionary conduct?
Conclusions:
Rambus’s challenged conduct is “exclusionary” in that it
Distorted an open competitive process by concealing (or misrepresenting) material information
Excluded equally efficient or superior alternative technologies
Entailed a conscious choice to jeopardize the enforceability of patented intellectual property
By distorting JEDEC’s standardization process, Rambus’s exclusionary conduct enabled it to obtain monopoly power
DX0129
Summary of ConclusionsSummary of Conclusions
4. Has Rambus’s conduct resulted in anticompetitive harm (actual or threatened)?
Conclusions:
Rambus’s conduct has caused significant anticompetitive harm by
substantially increasing prices (royalties) in the relevant technology markets
creating other actual / threatened distortions to competition in those markets (e.g., harm to innovation)
threatening to increase prices in downstream DRAM product markets
undermining confidence in open standards / standards processes
DX0130
Summary of ConclusionsSummary of Conclusions
5. What remedy (if any) is needed to restore competition / alleviate the anticompetitive effects of Rambus’s conduct?
Conclusions:
Rambus should be prohibited from enforcing against JEDEC-compliant DRAMs any patents filed (or based on filings) prior to June 18, 1996
This remedy should extend both to US and foreign patents
This remedy will restore competitive pricing in the relevant technology markets and mitigate other anticompetitive effects
DX0131
DRAM Industry OverviewDRAM Industry Overview
Technology Providers(e.g., Rambus, Jazio)
Consumers
DRAM Manufacturers
(e.g., Samsung, Micron)
Manufacturers ofDRAM-Related Logic
(e.g., Intel, AMD, ATI)
PC-OEMs, Server OEMs and other firms using DRAM(e.g., Dell, Gateway, HP, Cicso)
Technology
DRAM Chipsets, Graphics Cards
DRAM Upgrades Graphics Cards Upgrades
DX0132
DRAM BuyersDRAM Buyers
Personal Computers(62%)
Memory Modules(19%)
Workstation (5%)
Other: Printers, Routers, Facsimile... (4%)
Entry-level servers (4%)
Industrial & Additional Motherboards (3%)
Supercomputer,Mainframe, Midrange
(3%)
Percentage of DRAM Purchases, By Platform
Source: Gartner Dataquest, DRAM Supply/Demand Quarterly Statistics: Second Quarter 2000 Outlook, Bates No. HR905_06733 (August 14, 2000).
DX0133
Basic Economics of DRAM IndustryBasic Economics of DRAM Industry
Large capital requirements Economies of scale Interoperability Price sensitivity Commodity nature of DRAM
DX0134
High Cost of DRAM Fabrication PlantsHigh Cost of DRAM Fabrication Plants
DX0135
DRAM Chip Manufacturers TodayDRAM Chip Manufacturers Today
DX0136
DRAM Chip Manufacturers in the PastDRAM Chip Manufacturers in the Past
DX0137
Economics of DRAM ProductionEconomics of DRAM Production
High fixed costs Volatility / cyclicality Intense price competition Maximize capacity utilization / yield Intense cost cutting
DX0138
Reducing DRAM Production Reducing DRAM Production Costs / Increasing YieldsCosts / Increasing Yields
24/7 operation Clean rooms Extended equipment life Optimized production process Die shrinks Larger wafer size
DX0139
Economics of DRAM DemandEconomics of DRAM Demand
Multiple sourcing Long lead times Backwards compatibility Minimizing cost per bit Minimizing design, testing, and qualification
costs
DX0140
Evolution of DRAM StandardsEvolution of DRAM Standards
0
20
40
60
80
100
1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
Other FPM EDO SDRAM DDR RDRAM
Source: Cahners In-Stat “DRAM Memory Primer 2000” March 2000, Page 57 and In-Stat MDR “DRAM Market Forecast – Is there life After PCs” July 2002, Page 75 DX0141
Importance of Standards in Importance of Standards in DRAM IndustryDRAM Industry
Interoperability Diversity of use Common product design – DRAM producers Compatible design – DRAM users Facilitates price competition
DX0142
Nature of DRAM StandardsNature of DRAM Standards
Basic design specifications / protocols Focus on interface Parametrics Module standards
DX0143
How DRAM Standards Are SetHow DRAM Standards Are Set
Standard-setting organizations Private consortia Proprietary standards
DX0144
Economic Factors Influencing Success of Economic Factors Influencing Success of DRAM StandardsDRAM Standards
Open, consensus-based process Open availability of standard Royalties Implementation costs Manufacturing costs Evolutionary / revolutionary
DX0145
Economic Underpinnings of Economic Underpinnings of Evolutionary / RevolutionaryEvolutionary / Revolutionary
Reuse of existing knowledge / infrastructure Increasing marginal cost of changes Debugging / testing System-level design Risk
DX0146
JEDEC’s ProcessJEDEC’s Process
Diverse views / preferences Choice among alternatives Need for consensus Time to market Cost / performance considerations IP considerations “Satisficing”
DX0147
Problem of Hold-UpProblem of Hold-Up
“The hold-up problem arises because investmentsthat are specific to another party are vulnerable inrenegotiation – the other party can extract some or all of the value of the investments. The value of specific assets – those specific to a relationship with another party – are vulnerable to expropriation by that other party, because the assets have low or no value without the other party’s participation.”
P. McAfee. Competitive Solutions: The Strategist’s Toolkit (Princeton University Press, 2002, p. 365)
DX0148
Example of Specific InvestmentExample of Specific Investment
Potential Fuel Sources
• Coal
• Natural Gas
• Nuclear
• Solar
• Hydro
Power Plant Design
Ex Ante: Multiple Commercially Viable Fuel Sources
DX0149
Example of Specific InvestmentExample of Specific Investment
Potential Fuel Sources
• Coal
• Natural Gas
• Nuclear
• Solar
• Hydro
Power Plant Design
Ex Post: One Fuel Source
DX0150
Example of Specific InvestmentExample of Specific Investment
Mine 1 Mine 2
Mine 3Mine 4
Mine 5
DX0151
Example of Specific InvestmentExample of Specific Investment
Mine 1 Mine 2
Mine 3Mine 4
Mine 5$17/ton
$10/ton
$15/ton
$12/ton
$20/ton
DX0152
Example of Lock-InExample of Lock-In
Power Plant Locates Near Mine 1
Mine 1$10/ton
DX0153
Example of Hold-UpExample of Hold-Up
Power Plant Signs Coal Contract After Building
Mine 1
$20/ton
$10/ton
DX0154
Avoiding Hold-UpAvoiding Hold-Up
Power Plant Signs Coal Contract Before Building
Mine 1$10/ton
DX0155
Economic Literature on Hold-UpEconomic Literature on Hold-Up
Sanford Grossman and Oliver Hart, “The Costs and Benefits of Ownership: A Theory of Vertical and Lateral Integration,” Journal of Political Economy (1986)
Oliver Williamson, Markets and Hierarchy (Free Press, 1975)
Benjamin Klein, “Hold-Up Problem,” The New Palgrave Dictionary of Economics and the Law (Stockton Press, 1998)
DX0156
Application of Hold-Up to Application of Hold-Up to Standard SettingStandard Setting
Risk of hold-up depends upon characteristics of underlying industry / standards organization
Size of specific investments Cost of changing standards Importance of IP Ease of reaching agreement
DX0157
Application of Hold-Up to Application of Hold-Up to Standard SettingStandard Setting
Mechanisms for mitigating risk of hold-up ex ante
IP disclosure commitments
IP licensing commitments (e.g., RAND)
IP searches
DX0158
JEDEC: IP DisclosureJEDEC: IP Disclosure
Preference to avoid patents Early disclosure / good faith Disclosure applies to patents / patent
applications relevant to JEDEC standards / work RAND: mandatory for JEDEC; voluntary for
members Valid technical justification
DX0159
Application of Hold-Up to DRAM Application of Hold-Up to DRAM Standard SettingStandard Setting
Size of specific investments Substantial
Cost of changing standards Switching costs
Importance of IP High
Ease of reaching agreement Difficult and time consuming
DX0160
Standard Setting ProcessStandard Setting Process
A - C = Design / Technology Options
A
B
C
Standard Setting Process
DX0161
A
B
C
D
E
F
G
H
I
Standard Setting Process
Standard Setting ProcessStandard Setting Process
Feature 1
Feature 2
Feature 3
B, F, G
DX0162
A
C
Standard Setting ProcessB
DEF
Value of C
$
Standardization Confers ValueStandardization Confers Value
DX0163
Standard Setting Process
plant desig
n
compatibility
interoperability
manufacturing
$ $ $
Ex Ante Ex Post
time
Industry Commitment to DRAM StandardIndustry Commitment to DRAM Standard
DX0164
Standard Setting Process $ $ $
Ex Ante Ex Post
time
Ex AnteEx Ante Disclosure of IP Disclosure of IP
DX0165
C
A
A
C
Absent Patent Disclosure With Patent Disclosure
Ex AnteEx Ante IP Disclosure Can Alter the IP Disclosure Can Alter the Standard-Setting OutcomeStandard-Setting Outcome
DX0166
A
B Standard Setting Process
C
Optimal Choice Made with Full Optimal Choice Made with Full Information Early in the ProcessInformation Early in the Process
Ex Ante Disclosure of IP Rights
DX0167
A
C
Standard Setting ProcessB
Early Disclosure Leaves “Work-Around” Early Disclosure Leaves “Work-Around” Option Open Option Open
AC
C
A
DX0168
Standard Setting Process $ $ $
Ex Ante Ex Post
time
Ex Post Ex Post Disclosure of IPDisclosure of IP
DX0169
Standard Setting Process $ $ $
Ex Ante Ex Post
C
A
B
Patented Technology Standardized Patented Technology Standardized Absent DisclosureAbsent Disclosure
DX0170
Key Economic QuestionsKey Economic Questions
1. What are the relevant antitrust markets?
2. Does Rambus possess substantial market / monopoly power in such markets?
3. Did Rambus acquire market / monopoly power through exclusionary conduct?
4. Has Rambus’s conduct resulted in anticompetitive harm (actual or threatened)?
5. What remedy (if any) is needed to restore competition / alleviate the anticompetitive effects of Rambus’s conduct?
DX0123
Nature and Scope of Market DefinitionNature and Scope of Market Definition
Defines relevant scope of competitive activity Provides context to analysis of market power /
effects Common starting point for economic analysis in
antitrust-related inquiries
DX0171
Market Definition: MethodologyMarket Definition: Methodology
Analysis starts with market hypothesis Hypothesis tailored to antitrust issues presented Assume hypothetical monopolist Evaluate competitive constraints on hypothetical
monopolist Methodology is well-accepted
DX0172
Market Definition: MethodologyMarket Definition: Methodology
A
B D
C
E F
G H
DX0173
Technology MarketsTechnology Markets
Market for technology-related ideas / inventions Market definition concepts apply no differently
than in the case of physical “product” market Data on price / sales may be more limited Geographic scope is generally worldwide Well-accepted in economic analysis
DX0174
Technical FeasibilityTechnical Feasibility
Must be feasible to be in market Technical Issue
Testimony of engineers at trial Deposition testimony of engineers Testimony of Bruce Jacob Discussions with DRAM engineers
DX0175
Commercial ViabilityCommercial Viability
Commercially viable technologies are technologies that constrain prices of chosen technology
Parallel to the “SSNIP” test for markets with no price data Well-informed market participants treat as good
substitutes Serious consideration at JEDEC Qualitative judgments of knowledgeable engineers
DX0176
Commercial Viability ConsiderationsCommercial Viability Considerations
Time to market critical for JEDEC DRAM standards – “satisfice”
IP / royalties Cost of the solution to DRAM manufacturers
and others Performance benefits of the technology Strategic considerations that reflect the
competitive position of each member Every technology had problems to be solved
DX0177
Cost / Performance ConsiderationsCost / Performance Considerations
Different preferences on DRAM costs Some members more sensitive than others
Different preferences on DRAM performance Some members value performance more than others
Value of technologies depends on subsequent infrastructure AMD K-7 designed to exploit Burst Length of 8
Costs are uncertain until DRAM is manufactured commercially DDR in 1998 v. 2003
DX0178
F
E
D
C
B
A
Relevant Technology Market
Commercial Viability of TechnologiesCommercial Viability of Technologies
$ $ $ $
$ $ $
$ $
$
$
$
DX0179
Relevant TechnologiesRelevant Technologies
Programmable CAS latency Programmable burst length Dual-edge clock On-chip PLL/DLL
DX0180
Relevant Market Analysis: Relevant Market Analysis: Programmable CAS LatencyProgrammable CAS Latency
Technology used to set latency on DRAM Other “technically feasible” alternatives Some alternatives are “commercially viable” Market includes programmable CAS latency and
all commercially viable alternatives
DX0181
Technically Viable Alternatives for Technically Viable Alternatives for Programmable CAS LatencyProgrammable CAS Latency
Fixed CAS latency Set latency with pins Set latency with read command Program CAS latency by blowing fuses on the
DRAM Scale CAS latency with clock frequency
DX0182
Latency Technology MarketLatency Technology Market
Fixed CAS latency Presented at JEDEC
NEC Presentation at 42.3 Committee Meeting 76 (9/95)
Cost impact “A fixed DRAM is easier to test. Every time you add a new
feature to the DRAM, you have to test it. So, from a test cost point of view, there would have been an advantage to fixing every function in the DRAM, including programmable CAS and burst length.” Macri Trial Testimony at 4773
Potentially higher inventory costs to DRAM manufacturers – Macri Trial Testimony at 4763-64
DX0183
Latency Technology MarketLatency Technology Market
Programmable by pin strapping Presented at JEDEC
Micron Presentation at Special 42.3 Committee Meeting (7/00)
Cost impact “Q. So, one of the costs of these -- of this alternative would
have been to add extra pins on the DRAM? A. Yes, but if you're smart, you do it in a way where the cost is exceedingly minimal, and that's what, you know, we try to build the products on, being smart.” Macri Trial Testimony at 4767
DX0184
Latency Technology MarketLatency Technology Market
Programmable in read command Not presented at JEDEC
Mitsubishi Presentation at 42.3 Committee Meeting 60 (12/91) (Programming Burst Length)
Cost impact “The advantage would be that you would eliminate the mode
register and the circuitry required to decode special commands and put that information into the mode register, so it would make the part potentially smaller and simpler.” Jacob Trial Testimony at 5391-92
DX0185
Latency Technology MarketLatency Technology Market
Set by fuses Presented at JEDEC
Cray Presentation at 42.3 Committee Meeting 62 (5/92)
Cost impact “It would be potentially a simpler design. You would
eliminate the mode register. It would be potentially a smaller design and therefore a cheaper design. After blowing the fuse, you would only need to test one CAS latency value instead of having to test all possible CAS latency values, so it would be a cheaper alternative potentially.” Jacob Trial Testimony at 5382
DX0186
Commercially Viable Alternatives for Commercially Viable Alternatives for Programmable CAS LatencyProgrammable CAS Latency
Fixed CAS latency Set latency with pins Set latency with read command Program CAS latency by blowing fuses on the
DRAM Scale CAS latency with clock frequency
Latency Technology Market
DX0187
Relevant Market Analysis: Relevant Market Analysis: Programmable Burst LengthProgrammable Burst Length
Technology used to set amount of data read from a DRAM in response to a read request
Other “technically feasible” alternatives Some alternatives are “commercially viable” Market includes programmable burst length and
all commercially viable alternatives
DX0188
Technically Viable Alternatives for Technically Viable Alternatives for Programmable Burst LengthProgrammable Burst Length
Fixed burst length Use a separate pin or pins to set burst length Use burst-length-specific read commands to set
burst length Use a “burst terminate” command Program burst length by blowing fuses on the
DRAM
DX0189
Burst Length Technology MarketBurst Length Technology Market
Fixed burst length Presented at JEDEC
NEC Presentation at 42.3 Committee Meeting 76 (9/95) Cost impact
“A fixed DRAM is easier to test. Every time you add a new feature to the DRAM, you have to test it. So, from a test cost point of view, there would have been an advantage to fixing every function in the DRAM, including programmable CAS and burst length.” Macri Trial Testimony at 4773
DX0190
Burst Length Technology MarketBurst Length Technology Market
Programmable by pin strapping Not presented at JEDEC
Micron Presentation at Special 42.3 Committee Meeting (7/00) (Programming CAS latency)
Cost impact “The cost associated with each of those was relatively similar
in the large scheme of things, so I would say from a cost standpoint, that was a large factor in our decision.” Kellogg Trial Testimony at 5132
DX0191
Burst Length Technology MarketBurst Length Technology Market
Programmable in read command Presented at JEDEC
Mitsubishi Presentation at 42.3 Committee Meeting 60 (12/91)
Cost impact “Well, again, you would get rid of the mode register and
therefore the circuitry required to initialize it, which would make the part simpler to design and test and potentially cheaper to manufacture.” Jacob Trial Testimony at 5407-408
DX0192
Burst Length Technology MarketBurst Length Technology Market
Burst interrupt In SDRAM and DDR SDRAM standards and
proposed for DDR-2 Cost impact
“I mean, for DDR2, the DDR2 SDRAM standard, we do have burst interrupt, and it is fixed, and it's not a burden to the DRAM designers. The DRAM designers, if I recall their words, they said this is easy.” Macri Trial Testimony at 4775
DX0193
Commercially Viable Alternatives for Commercially Viable Alternatives for Programmable Burst LengthProgrammable Burst Length
Fixed burst length Use a separate pin or pins to set burst length Use burst-length-specific read commands to set
burst length Use a “burst terminate” command Program burst length by blowing fuses on the
DRAM
Burst Length Technology Market
DX0194
Relevant Market Analysis: Relevant Market Analysis: Dual-Edged ClockingDual-Edged Clocking
Technology used to increase the amount of data transmitted on the data bus
Other “technically feasible” alternatives Some alternatives are “commercially viable” Market includes dual-edged clocking and all
commercially viable alternatives
DX0195
Technically Feasible Alternatives for Technically Feasible Alternatives for Dual-Edge ClockDual-Edge Clock
Use two or more interleaved memory banks on-chip.
Keep each DRAM single data rate and interleave banks on the module (DIMM).
Increase the number of pins per DRAM Increase the number of pins per module Double the clock frequency Use simultaneous bidirectional I/O drivers Use toggle mode
DX0196
Data Acceleration Technology MarketData Acceleration Technology Market
Keep each DRAM single data rate and interleave banks on the module (DIMM) Presented at JEDEC
Kentron Presentation at 42.5 Committee Meeting 42 (12/99)
Cost impact More expensive due to components on module but used
cheaper DRAMs
Royalties may be a problem
DX0197
Data Acceleration Technology MarketData Acceleration Technology Market
Increase the number of pins per module Not presented at JEDEC Might not be commercially viable because of
increased costs of doubling the width of the data bus But data bus has been doubled in recent years by
Nvidia.
DX0198
Data Acceleration Technology MarketData Acceleration Technology Market
Double the clock frequency Presented at JEDEC
VLSI Presentation at 42.3 Committee Meeting 78 (3/96) Cost impact
“Well, a faster single edge clock has some enormous benefits in that we don't have to pay attention to this concept called duty cycle. Duty cycle -- you know, a clock has a pulse that is high and a pulse that is low, and the duty cycle is the length of the high pulse versus the length of the low pulse, and managing that is very difficult across real world conditions. It sounds simple; very complicated. Single edge clocking doesn't have that issue at all. So, that's a huge benefit to single edge clocking.” Macri Trial Testimony at 4779-4780
DX0199
Commercially Viable Alternatives for Commercially Viable Alternatives for Dual-Edge ClockDual-Edge Clock
Use two or more interleaved memory banks on-chip. Keep each DRAM single data rate and interleave banks
on the module (DIMM). Increase the number of pins per DRAM Increase the number of pins per module Double the clock frequency
Use simultaneous bidirectional I/O drivers
Use toggle mode
Data Acceleration Technology Market
DX0200
Relevant Market Analysis: Relevant Market Analysis: On-Chip DLLOn-Chip DLL
Technology used to synchronize the clock on the DRAM with the system clock
Other “technically feasible” alternatives Some alternatives are “commercially viable” Market includes on-chip DLL and all
commercially viable alternatives
DX0201
Technically Feasible Alternatives for Technically Feasible Alternatives for On-Chip PLL/DLLOn-Chip PLL/DLL
Put the DLL on the memory controller Use off-chip (on-module) DLLs Use a vernier method to account for skew Achieve high bandwidth using more DRAM pins,
not clock frequency Rely upon DQS data strobe to provide timing
DX0202
Clock Synch Technology MarketClock Synch Technology Market
Put the DLL on the memory controller Presented at JEDEC
Samsung Presentation at 42.3 Committee Meeting 78 (3/96)
Cost impact “You would eliminate the on-chip DLL, which would reduce the
power consumption of the DRAM. It would reduce the die size of the DRAM, which would reduce the manufacturing cost of the DRAM. You would reduce the testing costs of the DRAM because you don't have this PLL or, rather, this DLL that would be part of the DRAM. It would be a simpler design because it would not include a DLL and therefore cheaper, take less time. And it would cancel out more timing uncertainty than simply putting the DLL out on the DRAM itself, so you could potentially reach higher rates of speed than just using an on-chip DLL alone.” Jacob Trial Testimony at 5446-5447
DX0203
Clock Synch Technology MarketClock Synch Technology Market
PLL/DLL on module PLL on DIMM in registered DIMMs and in Kentron
QBM DIMM Cost impact
“You eliminate the on-chip DLL from the DRAM, thereby reducing its power consumption, reducing its cost, reducing the design time. … [Y]ou then move that design complexity onto a special DLL chip that goes onto the module, so you would be trading one for the other.” Jacob Trial Testimony at 5450
DX0204
Clock Synch Technology MarketClock Synch Technology Market
Vernier Presented at JEDEC
Synclink Presentation at 42.3 Committee Meeting 75 (5/95)
Cost impact “It's simpler to design than a DLL and it would cancel out
potentially more skew than a DLL so you could potentially achieve higher data rates using it. And burn less power.” Jacob Trial Testimony at 5452
DX0205
Clock Synch Technology MarketClock Synch Technology Market
No DLL at all Presented at JEDEC
SGI Presentation at 42.3 Interim Committee Meeting (7/97)
Cost impact “Q. Now, what, if any, would be the advantages of relying on
a DQS data strobe to provide timing rather than using on-chip DLLs?
A. Well, you would eliminate your DLL, which would make your design simpler. It would consume less power. The design would be smaller, cheaper to manufacture, and so forth.”
Jacob Trial Testimony at 5457
DX0206
Commercially Viable Alternatives for Commercially Viable Alternatives for On-Chip PLL/DLLOn-Chip PLL/DLL
Put the DLL on the memory controller Use off-chip (on-module) DLLs Use a Vernier method to account for skew Achieve high bandwidth using more DRAM pins,
not clock frequency Rely upon DQS data strobe to provide timing
Clock Synchronization Technology Market
DX0207
Asynchronous AlternativeAsynchronous Alternative
Asynchronous DRAM designs (e.g., Burst EDO) were an alternative to synchronous DRAMs
Some JEDEC participants (e.g., Micron) preferred asynchronous designs
Evolutionary advantages Choice of synchronous DRAM diverted
resources away from asynchronous designs
DX0208
Asynchronous – Burst EDOAsynchronous – Burst EDO
Alternative to both programmable CAS latency and burst length in SDRAM
Often presented at JEDEC NEC Presentation at 42.3 Committee Meeting (3/95)
Cost impact “It would have been a simpler transition because the technology
existed at the time. This was a technology that the engineers of the time were more familiar with. Asynchronous DRAM tended to have smaller die sizes like burst EDO at the time had a smaller die size than SDRAM and had better performance at the same speeds. So asynchronous potentially had better performance and cheaper implementation.” Jacob Trial Testimony at 5395-96
DX0209
Relevant Technology MarketsRelevant Technology Markets
Four relevant technology markets in this case:
Latency technology market Burst length technology market Clock synchronization technology market Accelerating data exchange technology market
These technology markets may be treated collectively as a single “cluster market”
Synchronous DRAM Technology MarketDX0210
A
B
C
D
E
F
G
H
I
Standard Setting Process
““Cluster” MarketCluster” Market
Feature 1
Feature 2
Feature 3
B, F, G
DX0211
Relevant Geographic MarketRelevant Geographic Market
For each relevant technology market, the relevant geographic market is the world
Buyers adopt technology with best value independently of geographic source
Technology licensed worldwide JEDEC-compliant DRAM produced and used
worldwide Negligible transportation costs
DX0212
Commercially Viable Alternatives for Commercially Viable Alternatives for Dual-Edge ClockDual-Edge Clock
Use two or more interleaved memory banks on-chip. Keep each DRAM single data rate and interleave banks
on the module (DIMM). Increase the number of pins per DRAM Increase the number of pins per module Double the clock frequency
Use simultaneous bidirectional I/O drivers
Use toggle mode
Data Acceleration Technology Market
DX0213
Data Acceleration Technology MarketData Acceleration Technology Market
Use toggle mode Considered at JEDEC by April 1992
“I think we had five companies showing what they called their own toggle mode in their presentations on the consideration of the first-generation synchronous DRAM.” Kelley Trial Testimony at 2585-86.
DX0214
Data Acceleration Technology MarketData Acceleration Technology Market
Keep each DRAM single data rate and interleave banks on the module (DIMM) Presented at JEDEC
Kentron Presentation at 42.5 Committee Meeting 42 (12/99)
Cost impact More expensive due to components on module but used
cheaper DRAMs
Royalties may be a problem
DX0215
Key Economic QuestionsKey Economic Questions
1. What are the relevant antitrust markets?
2. Does Rambus possess substantial market / monopoly power in such markets?
3. Did Rambus acquire market / monopoly power through exclusionary conduct?
4. Has Rambus’s conduct resulted in anticompetitive harm (actual or threatened)?
5. What remedy (if any) is needed to restore competition / alleviate the anticompetitive effects of Rambus’s conduct?
DX0123
Market and Monopoly PowerMarket and Monopoly Power
Market power is the ability profitably to maintain prices above a competitive level
Monopoly power refers to market power that is both substantial and durable
DX0216
Indicia of Rambus’s Monopoly PowerIndicia of Rambus’s Monopoly Power
In each relevant technology market, Rambus’s technologies today are the only commercially viable alternatives
Substantial barriers to entry
Ex post pricing of Rambus’s technologies substantially exceeds their ex ante value
DX0217
Development of JEDEC’s Development of JEDEC’s SDRAM StandardsSDRAM Standards
A
R
B
R
C
D
R
E
1993SDR
1999DDR
2003-04DDR-2
RR
DX0218
Evolution of DRAM StandardsEvolution of DRAM Standards
0
20
40
60
80
100
1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
Other FPM EDO SDRAM DDR RDRAM
Source: Cahners In-Stat “DRAM Memory Primer 2000” March 2000, Page 57 and In-Stat MDR “DRAM Market Forecast – Is there life After PCs” July 2002, Page 75
DX0141
DDRSDRAM
RDRAM Other
Types of DRAM Produced Today
DDRSDRAM
RDRAM
Types of DRAM Over Which Rambus Asserts Patents
Types of DRAM Over Which Rambus Types of DRAM Over Which Rambus Asserts PatentsAsserts Patents
Other
Source: Cahners In-Stat MDR “DRAM Market Forecast – Is there life After PCs” July 2002, Page 75 DX0219
Commercially Viable Alternatives Commercially Viable Alternatives
A
B D
C
E F
G H
Ex AnteEx Post
DX0220
Chipset
Motherboard
Mem
ory M
odule
Connectors
Gra
phics
Car
ds
Gra
phic
s Su
bsys
tem
s
Hard drive storage
CD
RO
M /
DVD
Driv
es
Modems
Net
wor
k C
ards
Sound Cards
Industry-Wide Coordination and Industry-Wide Coordination and Resource CommitmentResource Commitment
DRAM Standard
DX0221
Standard Setting Process $ $ $
Ex Ante Ex Post
R
A
B
Late Disclosure of Rambus Patents Late Disclosure of Rambus Patents Eliminated AlternativesEliminated Alternatives
DX0222
Costs of Changing Costs of Changing JEDEC Standards TodayJEDEC Standards Today
Develop new technology standards Difficulty of reaching consensus ex post Design, testing, and qualification of new DRAM
chips Redesign, testing, and qualification of existing
components Opportunity costs Cost of delay
DX0223
How Long Would It Take to Create a How Long Would It Take to Create a Non-Infringing Standard?Non-Infringing Standard?
19931991 1997
1995 1999 2002
1998 2003 +
SDR 2-6 years
DDR 4 years
DDR-2 5 years and counting
Volume Production
Volume Production
DX0224
Difficulty of Reaching Consensus Difficulty of Reaching Consensus Ex PostEx Post
Licensed producers have different incentives than unlicensed producers About 50% of the market has licenses
Users of specific features have distinct incentives AMD uses burst length of 8; Intel uses burst of 4
Greater disagreement ex post
DX0225
Barriers to EntryBarriers to Entry
Scale User switching costs
New product competes with existing devices Strong learning curve Sunk costs Patents
DX0226
Barriers to EntryBarriers to Entry
“The DRAM industry’s penchant for standardization combined with the Rambus marketing strategy of licensing all major vendors make it extremely unlikely that any potential competitor would be able to gain critical mass enough to challenge an already established and ubiquitous Rambus chip.”
Source: Rambus Business Plan Draft 6/26/89. CX0533, R114636
DX0227
RDRAMRDRAM Versus Versus DDRDDR Pricing Pricing
2.0
%
time
3.5
4.25
If Licensed at All
Samsung + Others
Hitachi
InfineonHynixMicron
?
DX0228
Key Economic QuestionsKey Economic Questions
1. What are the relevant antitrust markets?
2. Does Rambus possess substantial market / monopoly power in such markets?
3. Did Rambus acquire market / monopoly power through exclusionary conduct?
4. Has Rambus’s conduct resulted in anticompetitive harm (actual or threatened)?
5. What remedy (if any) is needed to restore competition / alleviate the anticompetitive effects of Rambus’s conduct?
DX0123
Exclusionary ConductExclusionary Conduct
What is exclusionary conduct?
Conduct that tends to exclude equal or superior products
Effect is anticompetitive – harms consumers
No valid efficiency rationale
DX0229
Exclusionary Conduct: Exclusionary Conduct: Major AssumptionsMajor Assumptions
Rambus possessed IP relevant to JEDEC standards / work
Rambus failed to disclose relevant IP as required by JEDEC rules / process
Rambus engaged in other, related misrepresentations while a member of JEDEC
After leaving JEDEC, Rambus continued to conceal its IP Before, during, and after JEDEC participation, Rambus
planned to enforce JEDEC-related IP Rambus was aware of legal risks associated with this
conduct (i.e., equitable estoppel)
DX0230
Reasons Why Rambus’s Challenged Reasons Why Rambus’s Challenged Conduct Is ExclusionaryConduct Is Exclusionary
Distorted JEDEC’s standard setting process by concealing (or misrepresenting) material information
Excluded alternative commercially viable DRAM technologies
Entailed a conscious choice to jeopardize the enforceability of patented intellectual property
DX0231
Exclusionary Conduct: Exclusionary Conduct: Misleading InformationMisleading Information
Concealing or providing misleading information is exclusionary when equal or superior products excluded
Concealing or providing misleading information prevents competition on the merits
Raising the perceived relative cost of alternatives is exclusionary
DX0232
““But-For” World AnalysisBut-For” World Analysis
“But-For” world = suppose (hypothetically) Rambus did not engage in challenged conduct
Compare actual and “but-for” world outcomes to determine effect of challenged conduct
Standard economic methodology
DX0233
If Rambus Had Disclosed IP to JEDEC:If Rambus Had Disclosed IP to JEDEC:No RAND LetterNo RAND Letter
Rambus documents state RAND not consistent with business model
Rambus wanted flexibility to charge different royalty rates
Rambus wanted RDRAM to succeed Not issuing RAND letter could have helped RDRAM Without RAND letter, JEDEC could not include IP in
standard
DX0234
JEDEC likely would not have included Rambus IP in standards Commercially viable alternatives Strong incentive to avoid running royalties because of
price sensitivity JEDEC process requires consensus
Incentive for ex ante negotiations RAND letter does not specify royalty rate Rambus had different incentives – “pure play”
technology company
If Rambus Had Disclosed IP to JEDEC:If Rambus Had Disclosed IP to JEDEC:RAND Letter IssuedRAND Letter Issued
DX0235
R1
B
R2
C
R3
D
R4
E
Rambus’s Conduct in Actual World:Rambus’s Conduct in Actual World:Outcome of JEDEC ProcessOutcome of JEDEC Process
SDR DDRR1, R2, R3, R4R1, R2
DX0236
R1
B
R2
C
R3
D
R4
E
SDR DDRB,C,D,EB,C
Rambus’s Conduct in “But-For” World:Rambus’s Conduct in “But-For” World:Outcome of JEDEC ProcessOutcome of JEDEC Process
DX0237
Exclusionary Conduct: Exclusionary Conduct: Rambus’s Costly InvestmentRambus’s Costly Investment
By not disclosing IP ex ante, Rambus knowingly incurred risk of having patents found unenforceable
Implication is that Rambus expected compensating benefits from non-disclosure
Like predatory pricing, this conduct is irrational absent expected benefits from excluding competition
DX0238
Rambus’s Conduct Would Not Be Rational Rambus’s Conduct Would Not Be Rational Absent Its Exclusionary EffectAbsent Its Exclusionary Effect
“…the only product that Rambus has about this is intellectual property. Doing anything as stupid as putting intellectual property in jeopardy by sitting in a meeting would have been ‑‑ passively sitting in a meeting, which is my understanding of what we did, would have been the stupidest management move that I could think of.
And you know, there isn't ‑‑ there is no rational motivation that I can think of that you would jeopardize the value of your patents by participating in a process that might deprive you of the right to enforce those patents.
I mean, it's ‑‑ there was very little to be gained and everything to be lost. I mean, that's not the kind of thing that you do with a rational manager.”
Deposition of Rambus Chairman, W. Davidow, 1/21/03, p.36 DX0239
Key Economic QuestionsKey Economic Questions
1. What are the relevant antitrust markets?
2. Does Rambus possess substantial market / monopoly power in such markets?
3. Did Rambus acquire market / monopoly power through exclusionary conduct?
4. Has Rambus’s conduct resulted in anticompetitive harm (actual or threatened)?
5. What remedy (if any) is needed to restore competition / alleviate the anticompetitive effects of Rambus’s conduct?
DX0123
Effects of Rambus’s ConductEffects of Rambus’s Conduct
Substantially higher, and discriminatory, prices (royalties) in the relevant technology markets
Actual (and threatened future) distortions to competition in those markets
Threat of higher DRAM prices Harm to standards setting processes
DX0240
DDRSDRAM
RDRAM Other
Types of DRAM Produced Today
DDRSDRAM
RDRAM
Types of DRAM Over Which Rambus Asserts Patents
Types of DRAM Over Which Rambus Types of DRAM Over Which Rambus Asserts PatentsAsserts Patents
Other
Source: Cahners In-Stat MDR “DRAM Market Forecast – Is there life After PCs” July 2002, Page 75 DX0219
Rambus RambusDRAM Mfrs. DRAM Mfrs.
Ex Ante Ex Post
Bargaining PowerBargaining Power
$ RAND or $0 Monopoly Royalty
DX0241
3.5%4.25%
?
SamsungOthers
Hitachi InfineonHynixMicron
Rambus’s Discriminatory DDR RoyaltiesRambus’s Discriminatory DDR Royalties
DX0242
Market Distortions Caused by Market Distortions Caused by Rambus’s ConductRambus’s Conduct
Harm to innovation Specific design investments Litigation costs Design-around costs Increased market uncertainty Potential delay in standard setting Long-run higher DRAM prices Harm to standards setting processes
DX0243
Consequence of Rambus’s Consequence of Rambus’s Changed PositionChanged Position
SDRAM / DDRRDRAM
RDRAM
.75 – 4.25% Royalty
1.5 – 2% Royalty
DX0244
Key Economic QuestionsKey Economic Questions
1. What are the relevant antitrust markets?
2. Does Rambus possess substantial market / monopoly power in such markets?
3. Did Rambus acquire market / monopoly power through exclusionary conduct?
4. Has Rambus’s conduct resulted in anticompetitive harm (actual or threatened)?
5. What remedy (if any) is needed to restore competition / alleviate the anticompetitive effects of Rambus’s conduct?
Economic Analysis of RemediesEconomic Analysis of Remedies
Preferred remedy: to restore competition to the situation that would have existed absent the anticompetitive conduct
As a practical matter, in this case the preferred remedy cannot be achieved Rambus’s monopoly power is durable The “but-for” world is now unattainable Some of the damage caused by Rambus’s conduct
can no longer be reversed
DX0245
Practical LimitationsPractical Limitations
An installed base of SDRAM and DDR SDRAM devices has been developed, and investment in JEDEC-compliant technology has already been committed
The DDR-2 standard has been developed largely under the same assumptions as were used for SDRAM and DDR SDRAM
Technological development in alternatives to Rambus’s claimed technologies has been foregone
DX0246
““Second-Best” Approach to RemediesSecond-Best” Approach to Remedies
If the conduct cannot be undone, then the effects of the conduct should be undone
The appropriate remedy to Rambus’s conduct thus involves minimizing the marketplace harm associated with the anticompetitive behavior
DX0247
Undoing the Effects of Undoing the Effects of Rambus’s ConductRambus’s Conduct
Rambus should be prohibited from enforcing against JEDEC-compliant DRAMs any patents filed (or based on filings) prior to June 18, 1996
This remedy should extend both to US and foreign patents
This remedy will restore competitive pricing in the relevant technology markets and mitigate other anticompetitive effects
DX0248