Retrospective - 2015-2025

The last decade will be remembered for a surge in interest in gallium oxide, the creation of the world’s first compound semiconductor cluster, the introduction of the VCSEL into smartphones, and the emergence of the GaN power electronics industry.

Two titans unite
2015. During the last 30 years, the two biggest sectors within our industry, LEDs and GaAs microelectronics, have headed in very different directions. In the LED industry five LED chipmakers rose to the fore in the late 1990s – Nichia, Cree, Lumileds, Toyoda Gosei and Osram – and held on to one of these top spots for many years. Now, though, the LED industry is far more fragmented. In stark contrast, the GaAs microelectronics industry has seen a dramatic contraction in the number of competitors, despite a growth in revenue over the last 30 years from below $0.5 billion to more than ten times this figure.

Big players that have left the GaAs industry include Anadigics, snapped up in 2016 by II-VI (now Coherent), which spent several years converting its RF line to the production of VCSELs. Other leading names have merged.

In 2002, Skyworks formed through the marriage of Alpha Industries and Conexant, and more recently Qorvo emerged from the union of two heavyweights, RFMD and TriQuint. The new entity started trading on 1 January, 2015.

That merger had much going for it. Many viewed the two companies as complementary, with RFMD having greater strength in RF technologies for handsets and TriQuint offering a better portfolio of defence products.

Investors clearly approved of the union, with the share prices of both firms enjoying double-digit jumps on disclosure of the plans. The new entity had the potential to initially net an annual revenue of just over $2 billion; and thanks to synergies providing yearly savings of $150 million, gross and operating margins were tipped to be 45 percent and 25 percent, respectively.

Ten years on, has the merger delivered on all those promises? Well, sales for the four most recent fiscal quarters have amassed $3.24 billion, with a gross margin around the targeted value.

However, the company is currently running at a small loss. This appears to be weighing on the share price, which has fallen from a peak of almost $200 in summer 2021 to around $70, the value of this stock on its first day of trading.

True heavyweight flexes its muscles
2016. If you are a regular reader of his magazine, then by the middle of the last decade you would have been aware of the tremendous promise of gallium oxide. With a whopping bandgap of around 5 eV – the precise value depends on the particular polytype – it has the potential to deliver a knockout blow to those two much-vaunted middleweights, SiC and GaN. Thanks to gallium oxide’s far wider bandgap, it should be able to reign supreme in that most valued of characteristics, the on-resistance as a function of breakdown voltage. But that’s not its only alluring attribute: it offers versatility, because thin films of gallium oxide can be deposited by a variety of techniques, including MBE and MOCVD; and there are several options for the substrate. Devices can be grown on sapphire or gallium oxide, which could tumble in price, because boules of this material can be grown by melt-based techniques.

Gallium oxide’s great promise has made it easy for researchers and developers to attract interest and initial investment. However, like any material, its future hinged on demonstrations of device capability. For gallium oxide, they came thick and fast in 2016.

During that year, two key breakthroughs emerged from a team at the Air Force Research Labs (ARFL). Using structures made from the b-polytype and grown on a native substrate, engineers working there grabbed the limelight by producing a MOSFET with a record-breaking critical field strength of 3.8 MV cm^-1 – that’s about four times the critical field strength of GaN. And later that year they realised another breakthrough, reporting the first ever enhancement-mode FET with a high-breakdown voltage in the off-state. This second success demonstrated that a lack of p-type carriers in gallium oxide is not a show-stopper to making a normally-off FET. The gallium oxide research programme at ARFL had been inspired by the work at Japan’s National Institute of Information and Communications (NICT). Researchers at NICT, working with domestic partners, had driven much early progress in materials and devices, and gone on to produce the world’s first MESFETs and MOSFETs in 2012.

Japan also provided another milestone in gallium oxide. In 2016, using a novel growth technique known as mist epitaxy, researchers at Flosfia reported β-Ga₂O₃-on-sapphire Schottky barrier diodes sporting an on-resistance that trumps that of state-of-the-art variants made from SiC. Several years ago, this University of Kyoto spin-off said that it would be commercialising its β-Ga₂O₃ power devices within a matter of months. The company failed to stick to its schedule, with a market launch postponed by reliability issues. However, it appears that Flosfia is now getting back on track, making progress, having recently announced that it has developed a prototype 600 V, 10 A junction barrier diode that will be sampled to customers this year. This device, with a turn-on of 0.9 V, incorporates p-type (IrGa)₂O₃. As well as this advance, the company has recently demonstrated the operation of a trench-gate MOSFET and scaled its production processes, establishing β-Ga₂O₃ growth technology for 4-inch wafers. With plans to also verify 4-inch wafer manufacturing this year, commercial prospects for Ga₂O₃ are on the up.

Creating the world’s first compound semiconductor cluster
2017. What is the most famous valley in the world? OK, it might not be the first response to trip off your tongue, but a great case can be made for Silicon Valley.

While this region may not have that many fabs today, the engineers that have worked here – many of them, graduates of Stanford University – are to be thanked for not only their advances in technology, but for giving all of us working in the semiconductor industry a higher status in society. Today technology is admired, benefitting many, even if the average man in the street has never heard of holes, bandgaps or lithography.

California is not the only part of the world with a cluster of semiconductor companies. They are also found in Leuven, Belgium; in Dresden, Germany; in Eindhoven, The Netherlands; and in Grenoble, France.

One of the strengths of all these clusters is that by having many companies on each other’s doorstep, often operating at different positions within the supply chain, they are able to support one another while attracting more engineering talent to the region. This helps to boost the kudos of the clusters, bringing in yet more business and innovation, while fostering an entrepreneurial outlook that gives local start-ups the best chance of thriving.

Efforts to create the world’s first compound semiconductor cluster can be traced back to 2011, when IQE’s founder, Drew Nelson, met the economic minister for Wales. Nelson put the case for a far stronger infrastructure within the UK to support the compound semiconductor industry. Back then, IQE had virtually no domestic customers. That’s not to say, however, that IQE was the only company in South Wales working in our industry: this part of the UK was and continues to be home to etching and deposition tool maker SPTS, now part of KLA; and a Microsemi facility with packaging expertise, now known as Microchip Technology Caldicot.

Nelson’s next steps included working within a group set up by the European Commissioner for Technology for the Digital Economy. Involved in a team considering key enabling technologies, he championed the construction of a sovereign capability for compound semiconductors in South Wales, to rebuild the continent’s manufacturing capability for this technology.

Building on these efforts, Nelson lobbied the Welsh Government and Cardiff University to set up an Institute for Compound Semiconductors. He had a vision for a local facility, delivering cutting-edge technology, developed using tools and processes compatible with high-volume manufacturing. In March 2015 the UK Government funded this initiative, and a year on planners gave the go-ahead for the construction of a new building for the Institute for Compound Semiconductors. Forming part of the Innovation Campus, the Institute opened in 2023 after delays partly associated with the pandemic, to play a leading role in the Translational Research Hub.

To help commercialise the technology developed at the Institute for Compound Semiconductors, in 2015 IQE and Cardiff University founded a joint venture: the Compound Semiconductor Centre. Using funding from Cardiff University and some equipment from CSC, it runs collaborative research projects.

In 2016, more links were added to the supply chain. In January, the UK government stumped up £50 million, funding the creation of a Compound Semiconductor Catapult, an open access R&D facility focused on helping UK businesses exploit advances in compound semiconductor technologies. And later that year, the UK’s Engineering and Physical Sciences Research Council poured £10 million into a Manufacturing Hub for Compound Semiconductors, targeting the translation of research into high-volume chip manufacturing.

With many pieces of the jigsaw now in place, in summer 2017 the cluster officially opened for business, taking the name CS Connected and holding events with speakers from all key stakeholders. The cluster also took a huge leap forward, grabbing an opportunity to get its hand on a production line. The 200 mm silicon fab in Newport, South Wales, had come into the hands of Infineon through its acquisition of International Rectifier. After the German powerhouse evaluated its global chip manufacturing capabilities, it viewed the Newport Fab as surplus to requirements. This facility was put up for sale, with Nelson leading a private equity buyout. Emerging from this deal, Newport Wafer Fab had guaranteed orders from Infineon for two years. Further ahead, those in the cluster hoped that the lines would also be used to produce compound semiconductor-on-silicon chips, aided by IQE’s acquisition of Translucent, a developer of rare earth oxides, which can provide a bridge between a silicon wafer and compound semiconductor epilayers.

Further advances within the cluster included the relocation of the CS Catapult to part of a very large building constructed for LG Semicon, with IQE taking the remainder, using it to create an epifoundry that could house up to 100 MOCVD tools. The cluster has also expanded to include Swansea University, which is constructing a Centre for Integrative Semiconductor Materials.

Threatening to derail some of this progress has been the ownership of the Newport facility, which, in 2021, was bought by Nexperia. As Nexperia is a Chinese-owned company, the take-over drew concern from the UK government, which spent many months deliberating over what to do. In November of 2022, the UK government finally came to a decision, ruling that the fab had to be sold. Twelve months on Vishay bought the facility, and late last year announced that it would be investing $51 million in the fab, which will diversify into the production of SiC power devices, and support the growth of the cluster.

Not helped by delays due to Covid, the Institute of Compound Semiconductors, part of the Translational Research Hub, opened in May 2023. Nobel Peace Prize winner and climate scientist Donald Wuebbles presided over the opening.

CSconnected, the world’s first compound semiconductor cluster, opened for business in 2017. By the end of the year this cluster, based in South Wales, had strengthened its capabilities, thanks to the launch of Newport Wafer fab, a facility Infineon deemed surplus to its requirements.

Getting the blues
2018. The VCSEL has a wonderful set of attributes. It is efficient, allowing it to run off a battery; it can be turned on and off at very high speeds, making it a great source for transmitting vast amounts of data; by adjusting the size of the aperture, it can produce single-mode emission with a circular profile, simplifying optics; and it is well-suited to high-volume manufacturing, partly because its design allows on-wafer testing.

However, there is room for improvement. Despites decades of development, the VCSEL spans a far narrower range of wavelengths than edge-emitting lasers and LEDs.

Expansion of the spectral domain has been very slow, given the long history of this device. Its roots go back as far as 1965, when Ivars Melngailis, working in the MIT Lincoln Lab, announced a ‘longitudinal injection laser’ emitting at 5.2 µm, formed from an InSb diode featuring polished top and bottom surfaces to ensure optical feedback. This device, an incredibly impressive feat for its time, is far from practical, with lasing requiring a hefty 20 A drive current and cooling to 10K.

The first real VCSELs came from the labs of Kenicha Iga from Tokyo Institute of Technology. In 1977, Iga proposed a design sharing many of the features of today’s VCSEL, and for the next 11 years he almost singlehandedly pioneered this class of device, before other groups noted his breakthrough and redirected their efforts towards this technology. Success followed, with research in the latter part of the twentieth century initially focusing on the near infrared, before attempts were made to widen the spectral range of the VCSEL in both directions. Initially, breakthroughs with the GaAs-based material system enabled 980 nm and 850 nm VCSELs, before emission stretched to around 650 nm. Wafer-fusion brought yet more success, allowing GaAs-based mirrors to be united with InP-based active regions to realise emission at 1.5 µm.

By the turn of the millennium, researchers started to consider the next goal: expanding emission to the blue and green. Success would open up new markets, allowing devices to be deployed for high-resolution printing, high-density optical data storage, chemical and biological sensing, full-colour displays and lighting.

Producing VCSELs operating in this spectral domain is far from easy. To reach these shorter wavelengths, the GaAs-based material system has to be replaced with one based on GaN and its related alloys. This switch may sound a simple, but it’s anything but.

The biggest issue is producing the mirrors that sit either side of the active region. This is relatively easy in a GaAs-based VCSEL, because this material system is blessed with the pairing of GaAs and AlGaAs. These III-Vs have very similar lattice constants, so strain is not an issue, and there is a significant difference in their refractive index, aiding reflection. When 20 pairs of alternating layers of GaAs and AlGaAs are used to make a mirror, it has a reflectivity of 99 percent, sufficient to make a high-performance VCSEL.

For blue and green VCSELs, to prevent strain from degrading the mirrors, GaN has to be paired with Al_0.83In_0.17N, a trickier alloy to grow that has a relatively low refractive index contrast. Growing two sets of GaN-based mirrors would take too long, so it is better to combine a nitride-based bottom mirror with a top one based on dielectrics, an approach pioneered by Nicholas Grandjean’s team at EPFL. Their high-point came in 2007, when they reported the first optically pumped GaN VCSEL.

Frustratingly, funding dried up for Grandjean, with those holding the purse strings arguing that the first electrical VCSEL would come from Japan. They were more or less right, and now we will never know if Europe could have been the trailblazer of this device.

The first electrically pumped GaN VCSEL actually came from National Chaio Tung University, Taiwan, announced in April 2008. This required cryogenic cooling, a restraint overcome later that year by Nichia, using a pair of dielectric mirrors. Nichia persisted with this design, increasing the output power to 0.62 mW in 2009. Over the next few years it continued to develop this device, but only realised minor additional gains.

Substantial progress came in 2018, when two Japanese companies, working independently, broke the 10 mW barrier – this is roughly the power required for augmented-reality devices, projection systems and displays. Stanley Electric, partnering with Tetsuya Takeuchi and his team from Meijo University, built on the work at EPFL, improving the growth conditions. In the autumn of that year, they reported a blue VCSEL with an output in excess of 15 mW. And in November 2018, Sony unveiled a 12 mW VCSEL with a novel design, featuring a far larger cavity that incorporates a thinned GaN substrate, and a curved mirror grown on the backside of the wafer.

Since then, gains have been modest, with the most powerful GaN VCSELs now emitting just over 20 mW.

Alongside the breakthroughs in blue VCSEL performance, 2018 will be remembered for the market success of infrared cousins. Up until then, datacoms provided by the primary revenue stream for this device. However, in 2018 the market for handsets caught up, thanks to Apple’s launch of the iPhone X. This smartphone featured two VCSELs: one for the dot projector, and another for the flood illuminator.

Many rivals in the smartphone sector also launched new models that incorporated VCSELs, used as the light source for time-of-flight technologies to avoid any potential IP issues. These sales have dwindled over the last few years, as makers of Android-based smartphones have tended to ditch the VCSEL in their latest models to free up space for the introduction of components for 5G networks. However, despite this loss of business, VCSEL revenue for the consumer electronic market is still increasing.

When Apple started to deploy VCSELs in its smartphones, Lumentum provided this device. But to reduce supply chain risk by adding a second source, in late 2017 Apple encouraged Finisar to become a VCSEL supplier, promising orders totalling $390 million, with the US manufacturer grabbing this opportunity. Over the intervening years Finisar’s VCSEL production business has been acquired by II-VI and subsequently incorporated into Coherent, which now has a significant share of the market.

For the last few years, Coherent and Lumentum have been competing for Apple’s business with Trumpf Photonic Components. Recently this chipmaker, which has shipped over 1 billion VCSELs to Apple, has been expandeing its manufacturing facilities for this device at its site in Ulm, Germany, which it bought from Philips Photonics in 2019.

Sales of VCSELs to the datacom market are now outpacing those for consumer electronics. Due to this, in 2022 Yole forecast that the datacom market would be overtaking that for consumer electronics. But more recent trends suggest that consumer electronics will remain the biggest market by far, accounting for $974 million in 2028, compared with $232 million for the telecom and infrastructure sector.

Several years ago, it looked like lidar for autonomous vehicles would be the next killer application for the VCSEL. However, the introduction of self-driving cars has been hampered by safety concerns, and here the VCSEL has come up against strong competition from fibre lasers and edge-emitting lasers, with the latter having the lion’s share of this market. This led Yole to predict that by 2028 VCSEL sales to the automotive and mobility market will net just $108 million, a relatively small slice of the $1.4 billion expected that year.

To secure a second source for VCSELs, Apple agreed to purchase $390 million of these devices from Finisar.

Facial recognition offers a lucrative market for the infrared VCSEL.

A phenomenal cash injection for silicon carbide
2019. In spring 2019, Cree, now Wolfspeed, re-aligned its business in emphatic fashion. It’s relatively new CEO Greg Lowe had no qualms in carving of the company’s Lighting Products division to Ideal Industries; giving the LED business that brought so many years of success a back seat; and focusing on wide bandgap materials, and RF and power devices. Helping to have made this monumental decision would have been a number of lucrative SiC wafer supply deals: contracts with Infineon, ST Microlectronics and other companies totalled $500 million.

By summer 2019 the company’s LED business had softened and it had netted the lion’s share of its $310 million sale to Ideal. These circumstances would have helped Lowe to move forward with his vision, investing $1 billion in massive expansion of SiC capacity. A state-of-the-art, 200 mm facility would be built by 2024, delivering a 30-fold expansion in capacity compared with the first fiscal quarter 2017.

Later that year the company had an offer it could not turn down – a $500 million grant from the state of New York. Instead of having to spend $450 million retrofitting existing infrastructure at its headquarters in Durham, NC, it could now invest $170 million in building a new, automotive-qualified 200 mm power and RF wafer fabrication plant in Marcy, New York. And instead of ramping up capacity by a factor of 30, it would go up 40 times.

Opened in April 2022 to much fanfare, Wolfspeed described this facility during its christening as ‘the first, largest and only 200 mm SiC fab in the world’. Back then, the company rode the crest of a wave, with many bold plans, plenty of backers, and a share price of over $100. But since then, the share price has tumbled, now trading well below $10, as the company has struggled from sluggish growth in the electric vehicle market, while facing increased competition from rival producers of SiC substrates, particularly from China. Last October Wolfspeed shelved plans to build the world’s most advanced SiC manufacturing facility in Saarland, Germany, and now it’s trimming its workforce by a fifth in a bid to cut costs. Amongst those redundancies is Greg Lowe, axed in November. The company is now seeking a new leader to transform its fortunes.

On 25 April, 2022, the day that Wolfspeed opened its 200 mm automotive-qualified SiC fab in Mohawk Valley, Upstate New York, the SiC tech supplier also announced a multi-year agreement with Lucid Motors, to supply SiC power semiconductors. These devices have been deployed in the luxury, all-electric Lucid Air.

Covid’s silver lining
2020. Who will forget 2020, a year that will go down in history as one when a pandemic rocked the world, with the rapid spread of Covid-19 having devasting consequences. As well as a high death toll – according the World Health Organisation, fatalities totalled more than 3 million – economies everywhere faced immense strain as the citizens of many nations were forced to isolate, sparking mental health issues with consequences lingering to this day.

While government rulings led many to work from home, those playing a hands-on role in semiconductor industries were seen in a different light. Compound semiconductor facilities continued to churn out chips around the clock, and even benefitted from opportunities associated within this global crisis.

To minimise loss of life, scientists devoted a great deal of effort to the speedy development and roll-out of a vaccine. And in addition to pursuing this game-changer, solutions were sought to kill this virus.

An attractive approach to dis-arming the virus is to subject it to emission from a deep-UV light source. The incumbent option, the mercury lamp, is bulky, fragile, requires a high operating voltage, and is plagued by environmental concerns. One way to address all these weaknesses is to switch to a battalion of deep-UV LEDs. Although they lag the performance of their blue-emitting cousins, they promise to come on in leaps and bounds through greater investment, spurred on by lucrative applications.

During the pandemic, interest in deep-UV LEDs skyrocketed, with firms launching new devices, alongside specialist growth tools and various disinfection systems.

Powerful deep-UV LEDs are far harder to produce than those emitting in the visible, due to a variety of issues, including realising sufficient light extraction and high enough levels of doping. Due to these and other challenges, the established makers of these devices were best-placed to ramp volumes, such as the partnership between Sensor Electronic Technology and Seoul Viosys, a subsidiary of Seoul Semiconductor. During the height of the pandemic, this collaboration started mass-producing UV LED modules that could sterilise 99.9 percent of the coronavirus within 3 seconds. In addition, they began developing a Photon Shower, described as a whole-body sterilisation solution that could sterilise germs on people’s clothing in a matter of seconds.

To support companies trying to manufacture the deep-UV LED, the Chinese equipment manufacturer AMEC launched a new MOCVD tool, the Prismo HiT3, said to be ideal for producing high-quality layers of AlN, a key material for this short-wavelength source.

During the height of lockdown, new disinfection systems employing deep-UV LEDs were also launched to market. Over the summer of 2020, water-treatment specialist AquiSense launched a surface disinfection system, and Singapore-based robotics manufacturer, Ostaw Digital, unveiled what was claimed to be the world’s first disinfection robot using LEDs emitting at appropriate wavelengths. When fully charged, this autonomous robot could be deployed for 5 hours, delivering a disinfection rate of 99.999 percent at a range of 2.5 m.

Sales of deep-UV LEDs emitting at wavelengths suitable for sterilising Covid-19 shot up during 2020, according to market analyst Yole, with revenue almost doubling year-over-year to $308 million. But that’s still well short of the value of the visible LED market, and even if the pandemic has provided a catalyst for healthy long-term growth of this source of deep-UV emission, this device is going to be valued more for its importance than its sales.