About thirty miles from the main Rethink office, a hive of advanced semiconductor activity is buzzing with activity. We were invited on a trip to explore this cluster of chipmaking wizardry, some of which is utterly foundational to the IoT. Of course, we could only scrape the surface, and for many IoT stakeholders, the intricacies of crystalline structures are utterly removed from their everyday business concerns, but in time, compound semiconductors are going to start cropping up in more and more devices, and powering more and more networks. As it is on our doorstep, it would almost be rude not to check out what was going on.
Broadly put, compound semiconductors have very different physical properties, due to their different materials. Electrons flow much more quickly through them than in pure silicon designs, which is why they are often used in light-based applications (photonics), and are capable of emitting light and microwaves, so are used in LED and microwave RF applications. Compound semiconductors also tend to have lower voltage requirements, and are more resistant to heat.
However, compound semiconductors are not a direct replacement for the silicon-based designs that power the integrated circuits inside all microprocessors. In terms of materials, silicon is used in the majority of semiconductors, with gallium arsenide (GaAs) conventionally thought to take second place, due to its use in lasers, power circuitry, microwave RF components, and solar panels.
The most common compound semiconductor materials are gallium nitride (GaN), gallium arsenide (GaAs), indium phosphide (InP), Silicon carbide (SiC), silicon germanium (SiGe), zinc sulphide (ZnS), and zinc selenide (ZnSe). The compounds are created by running an electric current through a substrate, and then filling a pressurized chamber with the desired gas – a process called Metal-Organic Chemical Vapor Deposition (MOCVD), which is using epitaxy to create the epitaxial film that forms the wafer.
This MOCVD process binds charged particles from the gas to the substrate. This process is done for each layer, with different recipes for each, culminating in a wafer that is usually 300-400 layers thick, each of which might only be two or three atoms thick. It is an intensive process that takes a lot of time, and requires very pure gases, to avoid introducing impurities. In terms of manpower, it is not very intensive at all, and can be heavily automated. The hard work is in the design process. Once you have built the wafer, you can then go about turning the wafer into the chips.
The compound semiconductor wafer creation process is very different from silicon, which essentially melts raw silicon and then grows it into layers of crystals that comprise an ingot, from which the wafers are cut and then polished. You then build the desired transistors on the wafers, before slicing the wafer into the individual dies that go on to create each processor or component. Here’s a good visualization of the process for silicon.
You can use compound semiconductors as the substrate, and then use MOCVD to build the epitaxial film on top of them, but there are some applications that use these epitaxial films on silicon substrates. Unsurprisingly, compound semiconductors are more expensive to create than silicon ones, and that is a reality that is decades away from changing.
So then, the trip itself. From the Rethink office in Bristol, it was only a short hop over to the IQE facility in Newport. The surrounding complex was actually first built as part of a government investment project that was meant to welcome LG Semiconductor to Wales, back in the early 2000s. However, the dotcom crash and South Korean government policy meant that LG swiftly exited the semiconductor business, ceding that market to Samsung, and so LG never arrived.
Since then, other businesses have moved into the surrounding space, but the main buildings that were constructed are now IQE’s facility, and what is apparently Europe’s largest data center – Next Generation Data. The building IQE now occupies was empty from construction until IQE got the keys in 2017. It is currently refurbishing and outfitting the facility with the 100 machines it will be using. Some twenty machines are on-site now, with ten of them active.
CEO Drew Nelson gave some context for the investment, saying that semiconductors are the driver for around 50% of global DFP, but that the progress of Moore’s Law is stalling because it is getting very hard to miniaturize transistors further. In terms of progress, Nelson said that if aviation had enjoyed the same advances as semiconductors, the transatlantic flight that cost $900 in 1971 and took seven hours would cost a penny and take around a second today. Compound semiconductors account for the top 10% of all semiconductor applications, he added.
Nelson says IQE has around a 60% market share in compound semiconductor wafers, supplying the chip makers with the materials needed to create the specialized chips. IQE’s designs are used in smartphones, cellular network infrastructure, data center and fiber optic interconnects, satellites, as well as defense-focused infrared imaging. The wireless segment accounts for the bulk of total revenue.
Apple’s FaceID system makes use of IQE’s products, using arrays of vertical-cavity surface-emitting lasers (VCSELs) to map a person’s face. With mobile payments rising, technologies like this become more valuable to phone makers. Nelson also said that 5G, particularly the mmWave frequencies, and LiDAR chips for self-driving vehicles, are going to be major drivers for compound semiconductor demand, as is infrared sensing in medical applications, where sensors could be used to measure blood sugar or toxicity levels.
The other main takeaway from the IQE presentation surrounded the opportunity for the British government, to create a sovereign capability to build these semiconductors, so many of which are used in applications of great strategic importance. This is the reason for the Welsh governments interest in promoting the collection of companies in the corridor of land that follows the major motorway through South Wales.
This cluster of companies is unique, according to Nelson, who said that this is the world’s first compound semiconductor cluster, and a chance to create a world-leading strategic capability. The UK government has established the Compound Semiconductor Applications Catapult to help drive investment, but the collection of companies and academic bodies in this informal cluster are planning on establishing CS Connected – a more official organization, to help drive investment in the region.
Nelson was asked why he though the UK could compete globally, when so much of the current semiconductor expertise is located in the US and China. His response was that because of the huge investments needed in processes and knowledge, UK manufacturing costs could undercut the overseas rivals. This is partly because the current supply chain has so many roundtrip inefficiencies (such as shipping IQE products to Taiwan, and then back to SPTS around the corner for the next step of the process, before going back overseas for assembly, and then finally being sold to BT as a finished box), that if the cluster moved to offering a more joined-up service, it could take advantage of the proximity and concentration of talent in the region to outbid international competition.
On the tour itself, Chris Meadows, IQE’s Corporate Systems Manager, expanded a little on this point. Meadows said that OEMs have started to engage with chipmakers much earlier in the process than they have historically done so. They are seeking more bespoke designs, and being involved in the earlier stages of manufacturer can ultimately lead to a better product. This shift is an opportunity for companies further down the supply chain, like IQE.
Meadows showed us one of the Aixtron machines that IQE is using in the new plant, which cost around $10mn each. They load the graphite substrates like a box of cassettes, with robotic arms manipulating them through the process, which takes a few hours. IQE has to purify the gases it purchases from suppliers, to remove impurities that could ruin the one-step process that has no room for error.
After the tour, we headed over to the Newport Wafer Fab, a facility that was initially built in 1982 by INMOS, before being bought by STMicroelectronics and later International Rectifier in 2002, before it was acquired by Infineon in 2015. A management buyout led to the creation of Newport Wafer Fab in 2017, where it buys the EPI wafers from the likes of IQE and then creates chips from them.
Once at the fab, we were treated to lunch, and then a round of speed-dating, where we met the rest of the participants in the CS Connected project – the aforementioned Catapult, Microsemi (component packaging, now part of Microchip), Cardiff University, Swansea University, SPTS (part of Orbotech that builds equipment used in chip-making), and the Compound Semiconductor Center (a joint-venture between IQE and Cardiff University, to bridge the research gap between the likes of IQE and the early-stage research done purely in academic environments).
A tour of the fab followed, led by NWF’s Sam Evans, Director of QM & External Affairs, in which we learned that in data center data transfer, switching from copper-based ethernet to light-based fiber (the realm of compound semiconductors), the data centers could enjoy a 70% reduction in power usage, simply by shifting from electrons to photons. NWF is building more facilities inside its fab, to scale production, and we were surprised to see how the components are handled inside large plastic boxes, a little like scaled-up versions of the security boxes used by retailers to prevent theft of Blu-ray boxsets, to keep them clean from impurities. Evans pointed out that it’s much easier to isolate the components from humans than to try and isolate humans from the components in the bulky space suits that most people envision when thinking about clean rooms.
Finally, the last session of the day allowed us to ask about the more practical functions of the cluster. This collective was likened to the plastic that holds together a six-pack of beer, but we wanted to know if there was any desire to create something more concrete – either by written contracts or vertical consolidation.
The consensus from the speakers was that in the ten to twenty year time frame, there might be, but right now, there’s too much innovation and not enough momentum behind any such push – ‘too many markets to create a monolith,’ as it was put by one speaker. There was a warning that monolithic economies of scale can’t be embraced in such a nascent industry, and that they collectively need to focus on developing the IP first and let the customers come later.
We pushed a little harder on the point of CS Connected being some form of industry or trade organization, and were told that it would act as a portal of sorts – a representation of a united front that evolve from an industrial body into an ‘entity of thought leadership.’ Fears were also expressed that scaling from the approximate 1,300 people involved in the current cluster, to the 5,000 additional jobs envisioned, could lead to the breakdown of the networking functions that have so closely bound the cluster’s members together.
That will be a tightrope that needs to be carefully crossed, but to our mind, this is an opportunity to create a very strong combined entity, built on an already market-leading foundation, supported by an academic pipeline that apparently can’t keep up with the deal volume of funding projects – but we couldn’t get any of them to say that they were in any great hurry to consolidate.