Globalfoundries to add another NVM to FDSOI
The company already offers magnetic RAM (MRAM) on its 22FDX process and is now researching other memory options, according to Gary Patton, chief technology officer.
The information was revealed on the side lines of the IMEC Technology Forum held in Antwerp last week as eeNews Europe interviewed Patton.
Patton was eager to emphasize that although foundry rivals such as TSMC and Samsung are pushing hard to introduce leading-edge 7nm processes connected intelligence is also about RF circuits and power efficiency. But is there a risk that Globalfoundries will fall behind in a race to the smallest geometries and start to lose customers?
Gary Patton, CTO of Globalfoundries.
“We’re all-in on the dual road-map strategy,” responded Patton.
That dual map strategy sees Globalfoundries first 7nm FinFET tape outs happening in the 2H18 with production in 2019 and further tape-outs in 2019, Patton said. Those early tape-outs are what Patton called “first-source” agreements. Patton said the first source customers include AMD, IBM and some other companies via an ASIC design platform.
It is important that Globalfoundries makes a success of its FinFET processes to keep its Fab 8 complex in Malta, New York busy and bring in the revenue to pay for R&D at smaller geometries such as 3nm.
Meanwhile on the other branch of its dual roadmap Globalfoundries is in competition with Samsung which offers 28nm fully depleted silicon on insulator (FDSOI) process and has announced an 18nm FDSOI follow on. “We’re very happy to see Samsung also participating. It only helps to keep more people in the market,’ said Patton. “We’ve got 36 design wins in FDSOI and tape outs on 22FDX.”
With 9 design wins in the US, 13 in Europe, 11 Asia and 3 in Japan, as you might expect there is a little more traction in Europe.
Next: Low risk
Patton said that FDSOI has turned a corner in terms of the perceived risk. “Now the roadmap is not an issue; IP availability not an issue and so the perception of risk has come down. Another measure is our FDXelerator partnership program launched with 7 partners in 3Q16 and now has 47. We’re heading towards 75 partners by the end of the year. We’ve had several real IC tape outs – rather than test chips – on 22FDX and we expect tape outs on 12FDX in 2020 with deliveries in 2021.”
Patton stressed that RF elements and millimeter wave circuits on FDSOI are world class making the process good for automobile radar applications. Indeed, its 22FDX process has just been certified to AEC-Q100 Grade 2 for production. As a part of the AEC-Q100 Grade 2 certification, devices must successfully withstand reliability stress tests for an extended period of time, over a wide temperature range.
Globalfoundries offers the AutoPro platform to help customers migrate their automotive microcontrollers and ASSPs to 22FDX and take advantage of capabilities in RF and mmWave alongside logic, non-volatile memory (NVM) and high voltage devices.
Globalfoundries has been a pioneer of embedded MRAM (see Globalfoundries offers embedded MRAM on 22nm FDSOI) but other companies are offering alternatives. Samsung is offering both embedded MRAM and flash options on its 28nm FDSOI (see FDSOI to get embedded MRAM, flash options at 28nm) and STMicroelectronics has opted for phase-change memory on 28nm FDSOI (see ST samples MCU with embedded phase-change memory).
Meanwhile TSMC is offering MRAM and a variant of ReRAM as an embedded memory on its 22nm FinFET process, according to reports (see Report: TSMC to offer embedded ReRAM in 2019).
Next: Globalfoundries and next NVM
“We are looking at other non-volatile memory options,” said Patton. Patton declined to say more about the options. It is notable that STMicroelectronics has chosen PCM (see Memory differences remain as ST chooses Globalfoundries for FDSOI) as well as chosing Globalfoundries as an FDSOI foundry partner but Patton would not be drawn. There are also a number of resistive RAM options being developed by fabless chip companies crying out for foundry partners. “22FDX will be a long-lived node so I expect will retrofit many technology modules,” said Patton.
With a change of subject Patton said there had been no change to plans to use extreme ultraviolet lithography with the 7nm process in 2019 (see Globalfoundries says EUV in production in 2019). “We have two EUV manufacturing tools in Malta, New York, with space for two more. We already have a third machine at Veldhoven [headquarters of ASML] being upgraded.”
Patton said: “We’ll launch with optical immersion lithography and then we will be able to offer performance enhancement to the 7nm technology.”
Patton repeated his exposition that in FinFET 14nm and 7nm are “major” nodes and he was glad that Globalfoundries had not put resources into 10nm. It is getting so expensive to engineer semiconductor manufacturing technology that it is only economic to migrate when you get the full 30 percent die area improvement, he said.
Patton said: “5nm is another half node and 3nm is likely to be the next full node.” He added that 5nm will likely be the last FinFET node and that 3nm would require significant transistor engineering. “It will be a big bill for a real node improvement,” he said. He added that nanosheets – flattened lateral nanowires each with a gate all around inside a fin – were a likely candidate for the 3nm process node. And this is something just announced by rival Samsung for their 3nm node coming in or after 2022 (see Samsung to introduce nanosheet transistors in 3nm node).
“People and technology miniaturization is moving slower. The introduction [of 3nm] will be three or four years after 7nm,” Patton concluded There was then the analogy that 7nm is an extreme sport and that will be like performing extreme sport at high altitude. “The air will be getting very thin.”
Related links and articles:
News articles:
Globalfoundries offers embedded MRAM on 22nm FDSOI
ST samples MCU with embedded phase-change memory
FDSOI to get embedded MRAM, flash options at 28nm
Report: TSMC to offer embedded ReRAM in 2019
Globalfoundries says EUV in production in 2019
Samsung to introduce nanosheet transistors in 3nm node
Memory differences remain as ST chooses Globalfoundries for FDSOI