
Intel works with 7 out of top 10 fabless companies, sees 18A test chip
Amongst its warning on lower income in Q4 and 2023, eeNews Europe highlights the seven out of the top ten foundry customers that Intel says it has increased its engagement. The company has taped out a test chip for its 18A process for a foundry customer this quarter.
While it won’t comment on the customers, other than Mediatek which is already public, the list of those engagements is reasonably clear. This is being driven by the risk of interruptions to supply from the US-China chip war and the worst case scenario of an invasion of Taiwan.
- Restructuring, slowing markets hit Intel results
- The drivers behind Intel’s Scottish software deal
- Financial figures show semiconductor market weakness
However the move to an entirely new foundry is not simple. Intel’s process technology and transistor design is very different to that of TSMC, the world’s largest foundry by far. Intel20A and 18A processes (2nm and 1.8nm equivalent) use a ‘ribbonFET’ transistor, and it is the 18A process that is aimed at external customers, said CEO Pat Gelsinger at the results. He says the company has taped out a 18A test chip for a company this quarter with several more taping out in coming months.
This will mean the target for Intel Foundry Services (IFS) will be for new chip designs in two to three years time, as fabs with the capability of 18A have yet to be constructed. Intel has previously said it epects to have the 20A node for volume production in 2024 and 18A in 2025 and Gelsinger said this is still on track. Porting existing designs from the TSMC process to IFS would require complex design and testing and so would take over a year as well.
“There are efficiencies for us to gain as we go to this foundry model,” said Gelsinger, including using more third party IP.
So who is planning to use IFS?
Nvidia, the second largest fabless chip designer, has joined a US defense project run by Intel to provide foundry services. This in essence pays for Nvidia to port its designs to the Intel 18A process.
Qualcomm, Broadcom and Marvell and Cirrus Logic are all US headquartered fabless semiconductors now that Broadcom has moved from Singapore and will see the political advantage of having indigenous US production in the IFS fabs in Ohio with leading edge processes.
Realtek, a Taiwanese PC and networking chip developer, is likely to follow MediaTek’s lead, giving seven out of the top ten fabless chip designers according to the latest data from Trendforce.
AMD is unlikely to place its faith in a direct competitor for production of its CPUs, GPUs and Xilinx FPGAs, going head to head with Intel in all three areas.
Novatek develops display drivers for LCD screens, and so the process technology requirements may not be a good fit for Intel. The same applies for Will Semiconductor, which makes DC-DC controller and power switches for high volume consumer power adapters and so won’t necessarily be able to use the Intel processes. Will is also based in Shanghai, and so will fall foul of the US-China trade war.
Intel is forecasting revenues around $64bn in 2023m down $4bn from previous guidance and down 16% on this year which is already a year of decline..
“We are still looking to have economic headwinds as we go into 2023,” said Gelsigner. The company is looking to cut costs by $3bn in 2023 and $8 to $10bn after that coming from the move to the foundry model. It is also reducing capital spending from $26bn to $21bn which will have an impact on the fabs it is building in Arizona, Ohio and Germany.
Related articles
- Opinion: Intel’s ‘smart capital’ is a warning from the past
- Intel confirms German fabs in €80bn European investment
- Intel boosts cloud EDA for its foundry business
- Intel installs first tools in Fab 34 in Ireland
Other articles on eeNews Europe
- Philips layoffs as supply chain, recall bites
- Renesas warns on falling semiconductor markets
- UK PM set to take on China with ‘NATO of technology’
- Samsung boosts AI in-memory processing with CXL
