3 ms·

There are three companies (TSMC, Samsung and Intel) that compete at the top end of chip fabrication in the whole world. One of them (Intel) didn't make chips fo

There are three companies (TSMC, Samsung and Intel) that compete at the top end of chip fabrication in the whole world. One of them (Intel) didn't make chips for anyone but themselves until about 4 months ago. Preparing a semiconductor for a given companies fab process takes time. You cant just call them up and ask for however many units of your design. Its a months to year long process of your engineers working with their engineers to optimize your circuit design to work best with their methods. Also there is only one supplier (ASML) in the entire world who makes a critical machine (extreme ultraviolet laser photolithography machine) needed for these cutting edge processes. It is unable to match the increased demand for their tools. And even if they were these super advanced might as well be magic machines have a lead time in years themselves as do the factories they go in to. If you want to start marking modern semiconductors or increase your existing capacity to do so it isn't as simple as throwing a couple million at it and having it running in a few months. Its billions of dollars in a massive sterile factory using machines made by a handful of suppliers taking up acres of land and consuming impressive amounts of electricity and water to do so. Semiconductor fabs are massive sprawling industrial campuses https://images.anandtech.com/doci/13749/intel_kiryat_678_678x452.jpg https://images.anandtech.com/doci/13749/intel_kiryat_678_678...

makapuf 5y ago
Many missing chips for automotive or devices are not gpu or top of the line level but simple microcontrollers. Those can be done with 130nm to 40nm process for the most modern. 130 nm has been used for pIII or amd athlon64 and dates from 2001. Those need near UV, not extreme UV.
paulmd 5y ago
> One of them (Intel) didn't make chips for anyone but themselves until about 4 months ago. actually iirc they've tried to get into the custom foundry business a couple times now (see: "Intel Custom Foundry" that got discontinued a couple years ago) but they never really attracted any major customers. https://semiwiki.com/semiconductor-manufacturers/intel/7912-intel-discontinues-the-custom-foundry-business/ https://semiwiki.com/semiconductor-manufacturers/intel/7912-... Part of this was due to the fact that Intel's design rules (the "rules" that govern chip design: how small wires and transistor features can be and so on) are completely different from every other foundry, so while porting from TSMC to Samsung or something like that certainly isn't easy, it's at least straightforward, whereas Intel's stack is basically totally different from everyone else's. One of the advantages Intel has is that the chip design and the process are tightly intertwined, if there is a problem with the chip design the designers can go down the hall and talk to the process people and see if they can come up with a solution at the process level. This obviously also became their downfall when 10nm slipped, and likely also it severely reduced their ability to sell their foundry services to other customers. According to that article they were apparently also rather secretive about their process (seeing it as part of their "special sauce" I guess) and I've heard many customers doubted Intel's commitment to the program as well as their neutrality. And I think the final piece of the puzzle was probably price - I'm guessing Intel was asking a royal sum compared to GF or Samsung or TSMC. After all... they're intel, and their process is the best! > One of the funniest stories I heard was about the first copy of the Intel 14nm design rules Atera got from Intel. They were heavily redacted, which is something I had never seen in the foundry business. After many delays Intel put their own implementation team on the first 14nm Altera tapeout and the result was a very competitive FPGA chip. If not for the continued delays, Xilinx would have been in serious trouble as the Intel 14nm FPGA, based on my experience with customers, beats the Xilinx 16nm in both density and performance. They are now trying to get back into the custom foundry business of course. Different management, and a very different business situation for Intel - if they can get other customers to help finance their node development that's a big win for them at this point.

makapuf 5y ago

Many missing chips for automotive or devices are not gpu or top of the line level but simple microcontrollers. Those can be done with 130nm to 40nm process for the most modern. 130 nm has been used for pIII or amd athlon64 and dates from 2001. Those need near UV, not extreme UV.

paulmd 5y ago

> One of them (Intel) didn't make chips for anyone but themselves until about 4 months ago. actually iirc they've tried to get into the custom foundry business a couple times now (see: "Intel Custom Foundry" that got discontinued a couple years ago) but they never really attracted any major customers. https://semiwiki.com/semiconductor-manufacturers/intel/7912-intel-discontinues-the-custom-foundry-business/ https://semiwiki.com/semiconductor-manufacturers/intel/7912-... Part of this was due to the fact that Intel's design rules (the "rules" that govern chip design: how small wires and transistor features can be and so on) are completely different from every other foundry, so while porting from TSMC to Samsung or something like that certainly isn't easy, it's at least straightforward, whereas Intel's stack is basically totally different from everyone else's. One of the advantages Intel has is that the chip design and the process are tightly intertwined, if there is a problem with the chip design the designers can go down the hall and talk to the process people and see if they can come up with a solution at the process level. This obviously also became their downfall when 10nm slipped, and likely also it severely reduced their ability to sell their foundry services to other customers. According to that article they were apparently also rather secretive about their process (seeing it as part of their "special sauce" I guess) and I've heard many customers doubted Intel's commitment to the program as well as their neutrality. And I think the final piece of the puzzle was probably price - I'm guessing Intel was asking a royal sum compared to GF or Samsung or TSMC. After all... they're intel, and their process is the best! > One of the funniest stories I heard was about the first copy of the Intel 14nm design rules Atera got from Intel. They were heavily redacted, which is something I had never seen in the foundry business. After many delays Intel put their own implementation team on the first 14nm Altera tapeout and the result was a very competitive FPGA chip. If not for the continued delays, Xilinx would have been in serious trouble as the Intel 14nm FPGA, based on my experience with customers, beats the Xilinx 16nm in both density and performance. They are now trying to get back into the custom foundry business of course. Different management, and a very different business situation for Intel - if they can get other customers to help finance their node development that's a big win for them at this point.