三星考虑将谷歌2nm TPU I/O Die后端设计外包给韩国DSP厂商

이전 기사보기
다음 기사보기
Samsung Weighs Outsourcing Back-End Design for Google's 2nm TPU I/O Die
Character Size Settings
가
The body content of the article will be changed to this letter size.
스크롤 이동 상태바
Semiconductor
ADTechnology, Gaonchips and Alphachips emerge as potential design partners amid rising foundry demand
Lee Jun
Published 2026.07.15 18:22
Reply 0
이 기사를 공유합니다
Google TPU. (Image: Google)
Samsung Electronics is considering outsourcing the back-end design of the input/output (I/O) die for Google's 2-nanometer tensor processing unit (TPU), as rising foundry demand stretches its internal engineering resources and prompts the company to seek support from external partners.
According to multiple industry sources on July 15, Samsung has recently approached its Design Solution Partners (DSPs) to assess interest in the back-end design work for the I/O die of Google's 10th-generation TPU, code-named Icefish, which will be manufactured on Samsung's 2nm process.
The TPU is Google's in-house AI accelerator used to power AI models including Gemini. Google is reportedly co-developing the 10th-generation TPU with MediaTek, with mass production expected to begin as early as 2028.
The TPU consists of a compute processor and a separate I/O die. The compute processor is expected to be manufactured by TSMC using its 1.4nm process, while Samsung will produce the I/O die on its 2nm node. The I/O die is responsible for transferring data between the compute processor and high-bandwidth memory (HBM).
DSPs adapt customers' chip designs so they can be manufactured at Samsung's foundries. The back-end design process includes physical implementation, such as placement and routing of logic circuits, design-for-test (DFT) implementation, and design verification. Samsung is considering outsourcing part or all of the work to DSPs, while also evaluating whether to keep portions of the project in-house.
Samsung handled the back-end design of Tesla's 2nm autonomous driving chip internally. However, industry sources said demand for Samsung's 2nm process has recently increased to the point where engineering resources have become constrained.
In addition to Google and Tesla, Samsung has reportedly secured Anthropic and DeepX as 2nm foundry customers. One industry source said some orders are shifting to Samsung because TSMC, the leading competitor in advanced-node manufacturing, is approaching capacity limits.
ADTechnology and Gaonchips are currently viewed as the leading candidates for the back-end design work. Both companies are already preparing or executing projects based on Samsung's 2nm process, the same technology node used for Google's TPU.
Industry sources said, however, that neither company has shown strong enthusiasm for the project. Both are already committed to major engagements, and back-end design services generally offer lower margins than full custom application-specific integrated circuit (ASIC) projects. Design houses typically prefer ASIC programs that cover the entire process from chip design through tape-out. Even so, the companies are expected to consider taking on part of the work to build a track record on a high-profile 2nm project for a global technology company.
Depending on the manufacturing process and contract structure, back-end design projects are typically valued at several billion won, while ASIC projects are generally worth tens of billions of won. For ASIC programs with guaranteed mass-production volumes, contract values can ultimately reach hundreds of billions of won.
ADTechnology is currently focused on its 2nm central processing unit (CPU) project, known as ADP620, with the goal of generating annual revenue exceeding 1 trillion won between 2028 and 2029.
Gaonchips is also preparing to participate in the approximately 800 billion won K-On-Device AI project led by South Korea's Ministry of Trade, Industry and Resources. The company plans to develop a 5nm advanced driver assistance system (ADAS) chip in collaboration with Hyundai Motor and other partners.
In addition to ADTechnology and Gaonchips, Alphachips has also been mentioned as a potential outsourcing partner. Unlike some of its rivals, the company reportedly views participation in the Google TPU project as a key growth opportunity and is actively pursuing the assignment.
2jun0504@thelec.kr
View other news
Copyright © The Elec Inc. No unauthorized transfer and redistribution
Comments
Name
Password
Enter comments
0
/ 400
Submit
Sort the comments.
Newest
Recommended
Top Comments
Best comment
It is automatically exposed by adding up the number of replies and recommendations.
close
더보기
Delete comments
Deleted comments cannot be recovered.
Do you still want to delete it?
Delete
Modify the comments
You can modify comments only within 1 minute after writing.
Text
Modify
My comment
Most Read
1
LG Display Says AI-Driven Manufacturing and R&D Save More Than 200 Billion Won Annually
2
Samsung SDI Supplies AI Data Center Battery Cells to Taiwan's Simplo Technology
3
Samsung Electro-Mechanics to Invest 23 Trillion Won in Sejong and Busan Through 2040
4
Samsung SDI to Invest 25 Trillion Won in Next-Generation Battery Production in Ulsan and Cheonan
5
​Samsung Reaffirms 1.4nm Mass Production in 2029, Sets SF1.4+ for 2030
6
US Media: Samsung and Anthropic Discuss Collaboration on 2nm AI Chips and Advanced Packaging
7
LG Energy Solution Begins ESS Battery Production at GM Joint Venture Plant
8
Samsung Foundry Showcases Future Production Pipeline at SAFE Forum
9
Is Meta's Cloud Plan Really a Sign of an AI Semiconductor Investment Bubble?
10
Synopex to Invest 15 Billion Won More in Vietnam Plant for EV FPCB Production
Editor’s Picks
PSK Holdings Wins Orders for More Than 300 Reflow Systems
SK hynix Expands Supplier Base for Hafnium High-k Precursors Used in DRAM
Samsung Electronics, Samsung Display Join Forces on Glass Interposer Technology
SK hynix Prices ADR Offering at $149, Raising Record $26.51 Billion
LG Energy Solution Expands Tesla Battery Production at Nanjing Plant
More Stories
LG Electronics Accelerates Push Into Robotic Actuator Business
LG Energy Solution Sees ESS as the Answer to AI-Driven Grid Bottlenecks
MIT Advances Toward KOSDAQ Listing After Securing Technology Evaluation Grade
Electrolytes Emerge as Key to Better, Lower-Cost Batteries