Source36氪•Stalecollected in 53m
Samsung to outsource Google TPU backend design
Strategic shift in AI chip manufacturing supply chain involving Google's TPU and Samsung foundry.
30-Second TL;DR
What Changed
Samsung outsourcing Google TPU backend design
Why It Matters
Outsourcing backend design allows Samsung to scale its foundry operations more effectively to meet the high demand for AI-specific hardware.
What To Do Next
Track the capacity of these design service firms as they become critical nodes in the AI hardware supply chain.
Who should care:Developers & AI Engineers
Key Points
- •Samsung outsourcing Google TPU backend design
- •Potential partners include AD Technology, Gaonchips, and Alphachips
- •Move driven by rising foundry capacity demands
- •Focus on optimizing semiconductor design service workflows
Deep Insight
AI-generated analysis for this event — not the original article.
Enhanced Key Takeaways
- •Samsung's strategy aligns with its 'Samsung Advanced Foundry Ecosystem' (SAFE) program, which aims to strengthen the design house network to compete more effectively with TSMC's Open Innovation Platform.
- •The shift toward outsourcing backend design (Physical Implementation) allows Samsung Foundry to reallocate internal engineering resources toward advanced node process development (2nm and below).
- •Google's TPU development has historically relied on a mix of internal design and external foundry partnerships, with this move signaling a deeper integration of Korean design houses into the hyperscaler's supply chain.
- •Backend design services involve complex tasks such as floorplanning, place-and-route (PnR), and timing closure, which are critical for the high-performance requirements of AI accelerators.
- •This outsourcing model is expected to reduce the 'time-to-market' for Google's custom silicon by leveraging the specialized expertise of firms like AD Technology and Gaonchips in Samsung's specific process design kits (PDKs).
Competitor Analysis
Ecosystem Model
- Samsung Foundry (w/ Design Houses)
- SAFE (Samsung Advanced Foundry Ecosystem)
- TSMC (w/ Design Partners)
- OIP (Open Innovation Platform)
- Intel Foundry (IFS)
- IFS Accelerator
Backend Focus
- Samsung Foundry (w/ Design Houses)
- Increasing reliance on 3rd party design houses
- TSMC (w/ Design Partners)
- Mature, extensive partner network
- Intel Foundry (IFS)
- Integrated IDM 2.0 model
AI Chip Focus
- Samsung Foundry (w/ Design Houses)
- High-volume TPU/ASIC production
- TSMC (w/ Design Partners)
- Dominant for H100/B200/TPU
- Intel Foundry (IFS)
- Emerging for custom silicon
| Feature | Samsung Foundry (w/ Design Houses) | TSMC (w/ Design Partners) | Intel Foundry (IFS) |
|---|---|---|---|
| Ecosystem Model | SAFE (Samsung Advanced Foundry Ecosystem) | OIP (Open Innovation Platform) | IFS Accelerator |
| Backend Focus | Increasing reliance on 3rd party design houses | Mature, extensive partner network | Integrated IDM 2.0 model |
| AI Chip Focus | High-volume TPU/ASIC production | Dominant for H100/B200/TPU | Emerging for custom silicon |
Technical Deep Dive
- Backend design (Physical Design) encompasses the transformation of a netlist into a GDSII file ready for manufacturing.
- Key processes include floorplanning to optimize signal integrity and power distribution across the TPU die.
- Timing closure is critical for TPUs, requiring precise management of clock trees to ensure high-frequency operation (often exceeding 2GHz).
- Design houses utilize Samsung's specific PDKs (Process Design Kits) to ensure the design adheres to the physical constraints of Samsung's FinFET or GAA (Gate-All-Around) nodes.
- The workflow involves sign-off verification using EDA tools (Synopsys/Cadence) to ensure the design meets thermal and power specifications before tape-out.
Future ImplicationsAI analysis grounded in cited sources
Samsung will increase its market share in the AI accelerator foundry segment.
By offloading backend tasks, Samsung can handle a higher volume of concurrent chip designs, improving its competitive positioning against TSMC.
Korean design houses will see a significant revenue increase by 2027.
Securing backend contracts for high-volume hyperscaler chips like Google's TPUs provides a stable and high-margin revenue stream for firms like AD Technology and Gaonchips.
Timeline
2018-02
Samsung launches the SAFE (Samsung Advanced Foundry Ecosystem) program.
2021-05
Google announces the TPU v4, marking a significant ramp-up in custom silicon production.
2023-06
Samsung Foundry Forum 2023 emphasizes the expansion of the design house partner network.
2024-04
Reports emerge of Google shifting more TPU production to Samsung's 5nm and 4nm nodes.
2025-11
Samsung announces advancements in 2nm GAA process technology, attracting interest from hyperscalers.
- 2018-02Samsung launches the SAFE (Samsung Advanced Foundry Ecosystem) program.
- 2021-05Google announces the TPU v4, marking a significant ramp-up in custom silicon production.
- 2023-06Samsung Foundry Forum 2023 emphasizes the expansion of the design house partner network.
- 2024-04Reports emerge of Google shifting more TPU production to Samsung's 5nm and 4nm nodes.
- 2025-11Samsung announces advancements in 2nm GAA process technology, attracting interest from hyperscalers.
Weekly AI Recap
Read this week's curated digest of top AI events →
AI-curated news aggregator. All content rights belong to original publishers.
Original source: 36氪 ↗
This is a summary, not the original. Read the source, or get the weekly briefing.
The weekly digest
One email a week. Unsubscribe anytime.