Intel’s long lasting schizophrenic relationship with cellular chips has ended with its decision to cease manufacturing 4G and 5G modems in the wireless wire area network (WWAN) business (see separate short story this week). Such chips went into notebooks, for which Intel will continue to make CPUs and rely on Taiwan’s MediaTek for modems in future.
The move appears to mark Intel’s judgment that anticipated volumes of 5G chips for notebooks are too low to justify continued investment, as wireless connectivity itself becomes commoditized and the added value resides in chips or accelerators for the applications.
But while it may reflect Intel’s ultimate failure to establish a stronghold in the wireless semiconductor market it also highlights the changing dynamics of the global semiconductor industry. The latter is undergoing extensive rationalization amid a drive to reduce dependence on one or two foundries for manufacture of the most advanced chips at the densest processes of 3nm and below, measures that equate roughly to the size of transistors.
Intel’s own checkered story in wireless dates back at least to 2007 when it decided not to make chips for Apple’s first iPhone. Although the company conceded this was a mistake with hindsight given the iPhone’s success, Intel remained ambivalent and nine years later in 2016 pulled out of the 4G smartphone chip market altogether, deciding to focus its mobile strategy on the then impending 5G field where it hoped to target emerging vertical markets for more complete subsystems, in automotive, robotics ad drones for example. At that point Intel cancelled its Atom line of smartphone chips then under development, losing confidence in its ability to compete with the likes of Qualcomm, Samsung, and increasingly Apple with their chips using technology licensed from Arm.
Arm had come to dominate chip architectures for mobile devices, especially smartphones, supplying either full CPUs or instruction sets that other semiconductor companies could use to design their own, but without having to reinvent the fundamental operations of the processor. This means that any Arm-compliant CPU should execute given code in the same fundamental way, even if some CPUs are better optimized for specified tasks than others.
Intel’s next retreat came in mid-2019 when its 4G and 5G modem business was sold to Apple, which has since taken the lead for integrated smartphone system on chips (SoCs) that combine the CPU with other essential components such as the cellular connectivity. Such integration enables the optimization of performance and power efficiency essential for smartphones and tablets, but not for notebooks where there is the luxury of more space for separate components on the board, as well as for larger batteries.
At that stage Intel retained various 4G and 5G patents and continued making modems for notebooks where they were deployed separately from the CPU on the boards as options. But now Intel has sold the remaining modem business to MediaTek and plans to complete the associated technology transfer by May 2023. After that it will retain a small team to support joint customers. Intel’s OEM partners will be able to continue collaborations with MediaTek to provide updates and upgrades to existing products. This was curious in the sense that just a month earlier ahead of MWC 2023, Intel had issued a press release trumpeting accelerated 5G leadership with new products, but was actually talking only about the core and virtualized RAN functions.
At any rate, Intel has now issued a statement to the effect that the exit from LTE and 5G based WWAN will allow the company to focus on its IDM (Integrated Device Manufacturing) 2.0 strategy. The reference to 2.0 implying a second generation is rather arbitrary here in Intel’s context but is loosely associated with broader changes across the global semiconductor ecosystem. Chip making began by being vertically integrated with a few large manufacturers such as Intel designing as well as fabricating their own silicon, as well as being responsible for their own instruction sets.
But as chip manufacturing advanced to ever greater transistor densities, fabrication became increasingly specialized and expensive, contracting down to an ever smaller group of foundries for the more advanced state of the art processes at the smallest transistor dimensions. At the same time, the advanced foundries themselves became dependent on advanced lithographic processes for printing integrated circuit patterns onto silicon wafers. It was here that the USA has been able to hurt China’s chip industry by imposing extraterritorial restrictions on all exports containing a significant amount of its technology through the Foreign Direct Product Rule. China had failed to invest significantly in this area and became dependent on overseas foundries, such as Taiwan’s TSMC, which in turn used lithographic machines from these other companies.
While established chip makers such as Intel continued to fabricate some of the less advanced chips they became increasingly dependent on the leading foundries for the cutting edge silicon as required by the latest smartphones. We were in the era of fabless semiconductors whereby production was often outsourced to a third party foundry.
Qualcomm, Nvidia, Broadcom and also Intel’s CPU rival AMD are all fabless, as is Apple which has relied on TSMC for its chip fabrication. However, Intel itself continued to qualify as an IDM with its own inhouse manufacturing capability, even if this was only for older generations of process design. Samsung is also an IDM, as is Texas Instruments after establishing a foundry in China in 2010.
There are also pure play foundries, notably GlobalFoundries with plants in Singapore, Germany and the USA, and Taiwan’s UMC, besides global leader TSMC. Given this relative concentration in the Far East and especially Taiwan, there have been moves to diversify foundry capability geographically, as well as by leading nations to acquire their own plants at a time of growing global geopolitical uncertainty. Intel alone of the major chip companies has moved to reacquire advanced foundry capability, as well as establish a base in the USA, to reduce dependence on third party foundries and in Taiwan especially in the light of increasing risk of a Chinese invasion.
Intel made its first major move on this front in March 2021 when it unveiled plans to invest about $20 bn on two chip fabs in Ocotillo, Arizona, with plans to dedicate at least a portion of the output to a new subsidiary to be called Intel Foundry Services. This new chip manufacturing arm would produce silicon for its x86 architecture, as well as Arm designs like those used by Apple. Indeed, Intel harbors hopes of regaining Apple as a customer at the foundry level, having lost it years earlier as the latter moved into its own fabless production.
With such investments, global semiconductor supply chains are becoming increasingly fluid and entwined, such that major players can be IDMs for some designs, fabless for others, and yet can act solely as foundries. Companies can be competitors, customers and suppliers at the same time, and this can lead to some tensions, as between Intel and TSMC.
Intel has become dependent on TSMC for advanced processes in order to avoid losing ground to fabless rivals, notably AMD, which had no qualms about relying on external foundries. Intel adopted a middle way by stitching chip subcomponents called tiles made by TSMC together in its own foundries so that it could at least exploit its packaging technology. Intel was planning to use TSMC’s 3nm process based GPU tiles for its forthcoming Arrow Lake designed to enter desk top systems in 2024. This would be a multi chiplet design featuring a CPU it would fabricate itself, alongside the TSMC GPU tile.
But after its new foundry capability come fully on stream in the USA, Intel plans to become not just more self-sufficient for its own chips but also a provider of fabrication capability to others in competition with TSMC. Meanwhile it looks like radio chips for the RAN and smartphones will continue to be produced by fabless companies, so it is possible Intel’s expanded foundry capability will come into play there.
The new IFS division announced in January 2023 that it had won an order from a major “cloud, edge, and datacenter solutions provider” for producing chips based on its Intel 3 production node. This added to customers such as MediaTek to accumulate a lifetime deal value for IFS so far of over $4bn, according to CEO Pat Gelsinger.