AI summary by TwinMind

Lessons from Intel's First Foundry

Asianometry

37m 0s

Intel's Crown Jewels and Strategic Motivation (2000s)

  • Intel identified two crown jewels: leading-edge manufacturing tech and x86 architecture
  • Manufacturing leadership: 1-2 generation lead over competitors through technical choices
    • Gate Last vs Gate First method (backed by IBM)
    • TSMC initially pursued Gate First before abandoning
  • Rising fab costs threatened sustainability with PC market revenues alone
  • Strategic response: aggregate external customer demand to fund next nodes
  • Paul Otellini sold XScale to Marvell (2006) to focus on x86
  • Otellini's 2007 Financial Times interview: "entire internet coded for x86 base"
  • iPhone proved x86 mobile strategy incorrect, rushed Intel Atom development

Custom Intel Architecture Foundry Launch (2008)

  • Original name: Custom Intel Architecture Foundry (CIAF)
  • Goal: monetize manufacturing + defend Intel Architecture franchise
  • Small team formed for "turnkey custom foundry services"
  • Initial VP left, replaced by Sunit Rikhi (Technology & Manufacturing VP)
  • Rikhi served as general manager for 7 years

Building Foundry Capabilities

  • Intel's processes optimized for x86 CPUs only, not external customers
  • Pizza kitchen analogy: specialized for "pepperoni" (x86), needed flexibility for various "pizza types"
  • Required extensive work to match TSMC/UMC/GlobalFoundries capabilities
  • Needed complementary IP ecosystem (analogous to pizza toppings)
  • TSMC had 3+ decades of foundry experience vs Intel's startup effort

Abandoning Intel Architecture Focus (2009)

  • Parallel effort: Intel-TSMC alliance for Atom IP licensing
  • Both initiatives failed due to low customer demand
  • Atom ecosystem immaturity vs established ARM platform
  • Otellini reluctantly dropped "Intel Architecture" from name (2009)
  • Became "Intel Custom Foundry" (ICF)
  • Otellini maintained belief ICF should focus on Intel Architecture

Competitive Constraints

  • Major issue: competing with potential customers (AMD, Nvidia, Qualcomm)
  • TSMC's charter: never compete with customers
  • Intel chose slow, discreet rollout
  • Restricted to non-competitive areas only
  • Focused on Field-Programmable Gate Arrays (FPGAs)
  • CFO Stacy Smith (2010): "not building broad-based foundry"

Early Customer Acquisition

  • Target customers: startups tired of TSMC dependence
  • Structured contracts with severe IP protection penalties
  • First Process Design Kits: 32nm, quickly switched to 22nm
  • First customer: Achronix (FPGA startup, Oct 2010)
    • 1% capacity allocation of 22nm FinFET
    • CEO: "Intel far ahead of anyone else in new process generations"
  • Additional customers: Tabula (Feb 2012), Netronome (Apr 2012)
  • All customers previously used TSMC nodes

22nm FinFET Technology Leadership

  • Intel's 22nm with Tri-Gate FinFETs (late 2011)
  • Decade of development to master FinFET manufacturing
  • Ivy Bridge CPUs: scheduled Apr 2012, shipped June 2012
  • Achronix Speedster22i FPGAs: volume production early 2013
    • 50% power consumption vs competitors (Altera, Xilinx)
    • Competitors limited to TSMC 28nm (volume-constrained)
    • Intel 1-2 generations ahead

Design-Manufacturing Integration Strategy

  • Mark Bohr (Oct 2012): "traditional foundry model running into problems"
  • Intel's IDM advantage: design and process development under one roof
  • Full service offering: wafer fab + design + advanced packaging
  • Embedded Multi-die Interconnect Bridge: Intel innovation for foundry customers
  • Reality: internal kingdoms struggled to collaborate
  • Technology group resistance to customer feedback
  • Product group worried about competitive comparisons

Internal Organizational Challenges

  • Many Intel employees viewed ICF as waste of money/distraction
  • Product group concerns about external CPU performance comparisons
  • Technology group initially supportive, then withdrew support
  • ICF struggled to get internal cooperation for customer requests
  • Technology R&D extremely secretive, refused customer insights
  • Resisted transistor modifications along power-performance curve
  • Mark Bohr brought in to leverage senior management for R&D cooperation

Major Customer Win: Altera (2013)

  • FPGA market dynamics: Altera (TSMC customer) vs Xilinx (UMC customer)
  • UMC decline: 65nm node issues delayed Xilinx products
  • Xilinx defected to Samsung + TSMC (Feb 2010)
  • TSMC gave Xilinx 28nm access to win over Samsung
  • Altera needed technical advantage, turned to Intel
  • Altera deal: 14nm node production, trailing edge at TSMC
  • CEO John Daane: "Intel's 14nm is second generation FinFET"
  • TSMC shock: Morris Chang earnings call (Apr 2013)
    • "I really hate to lose even part of old customer"
    • "Intel very serious competitor to our customers"

14nm Node Challenges and Expansion

  • Additional customers: Panasonic, LG for system-on-chips
  • Altera design delay: resources consumed by TSMC 20nm Arria 10
  • Intel 14nm node: 1 year late due to aggressive technology definition
  • Originally scheduled 2013, reached high volume mid-late 2014
  • Missed 2-year Moore's Law cadence
  • Delay allowed TSMC/GlobalFoundries-Samsung to close gap
  • Final node: second-gen FinFET, denser than TSMC 16nm first-gen

Apple Negotiations and 10nm Disaster

  • Apple-Intel meetings (2013) for A9 dual-sourcing
  • Apple rejected 14nm, interested in 10nm (due 2016)
  • 10nm would have been 2 years ahead of TSMC N7
  • A9 deal went to Samsung instead
  • Intel 10nm problems: 50% area shrink, 2.7x transistor density increase
  • Parallel development teams: 14nm issues discovered too late for 10nm
  • Cobalt deposition tools not ready, EUV unavailable until 2019
  • "50-story skyscraper with inadequate foundations" - complete restart needed
  • 10nm delays (2015) dried up foundry business backlog
  • 10nm Intel chips finally shipped 2019, foundry already dissolved

The Fatal Altera Acquisition (2015)

  • Intel acquired Altera for $16.7 billion (June 2015)
  • CEO Brian Krzanich strategy: integrate FPGAs into data center CPUs
  • Goal: offload workloads to FPGAs instead of GPUs
  • Integration work too complex, strategic product never shipped
  • Disaster for ICF: acquiring marquee customer sent bad message
  • Potential customers (Qualcomm) feared Intel would acquire them too
  • Reinforced Morris Chang's competition concerns

Dissolution and End (2015-2018)

  • Combination: Altera acquisition + 10nm failure = crippling
  • Sunit Rikhi retirement (2015) after 8 years as GM
  • Replacement favored Technology group over customer advocacy
  • Customers worked directly with Technology, cutting out ICF
  • Division employment: 1,100+ people worldwide at peak
  • Quiet dissolution by 2018
  • Book of business: 10+ customers, $1 billion+ contracted revenue

Lessons for Intel Foundry Services

  • Pat Gelsinger mistake: rebuilding from scratch vs learning from ICF
  • ICF achievements: real competitive threat to TSMC per Morris Chang
  • Key takeaways:
    • Organizational support needed across entire company
    • Foundry needs autonomy to modify technology
    • Leading-edge node is primary customer draw
    • 18A/14A must work for foundry customers, not just CPUs
    • Long-term commitment required (10+ year endeavor)
    • Humility essential: ICF's ragtag nature was strength
    • Study Morris Chang's foundry business philosophy

TwinMind

Get summaries like this for your own meetings

TwinMind takes notes in the background during meetings, lectures, and calls, then writes the summary for you.

Free plan to get started

No bot joins your calls

Phone, desktop & Chrome apps

Your next meeting, summarized like this

TwinMind listens in the background and turns meetings, lectures, and calls into searchable notes you can chat with. Start free.

Available on phone, desktop & Chrome

Your next meeting, summarized like this

TwinMind listens in the background and turns meetings, lectures, and calls into searchable notes you can chat with. Start free.

Available on phone, desktop & Chrome

Your next meeting, summarized like this

TwinMind listens in the background and turns meetings, lectures, and calls into searchable notes you can chat with. Start free.

Available on phone, desktop & Chrome