Today the LEDs emitting in the blue and green are made from the nitrides, while arsenides enable the construction of the orange, yellow and red emitting cousins. 

Trimming the number of material systems from these two to just one would be better on multiple fronts. It would simplify packaging; it would allow driving systems to be optimised for specific electrical and material characteristics; and it would open the door to the production of full-colour LED displays and white-lighting systems based on a single chip.

When exploring this route, there is no question which material system should be used: it has to be the nitrides, because the arsenides can’t stretch to shorter wavelengths, due to a switch to an indirect bandgap. Ditching arsenide devices would also lead to other benefits, because their efficiency plummets at higher temperatures, and their wavelength shift with current exceeds that for nitrides.

Efforts to extend nitride LEDs from the green to yellow, orange and red are already well underway. Success is not easy, however, as quantum wells must be rich in indium to reach these wavelengths – and that impairs material quality, and ultimately efficiency. Making matters worse, internal electric fields increase in strength at longer wavelengths, pulling electrons and holes further apart and hampering radiative recombination. 

One way to improve performance is to reduce the strain in these devices. For example, when a team from Ostendo Technologies inserted a compliance layer beneath the active region, they produced the world’s first GaN LED delivering tuneable, full-colour emission. This chip features intermediate carrier-blocking layers to control electron and hole transport between the quantum wells emitting in the red, green and blue (see p. 44). 

Ostendo is now developing its technology for the display market. The promise of replacing backlit screens with those based on LED pixels is very attractive, as it could lead to simpler, more robust displays with greater efficiency.

Researchers in Japan are also developing long-wavelength nitride LEDs. A partnership between scientists at Tokyo University of Science and Meijo University has developed a hybrid LED featuring an orange-emitting quantum well stacked on a blue one. The latter does not actually emit, instead serving as a strain releasing layer (see p. 58).

For both teams, emission efficiency must increase at longer wavelengths. That’s not surprising, given that this technology is in its infancy. But hopefully great advances will be made, leading to widespread commercial adoption of full-colour GaN LEDs.

Macom, a US company that makes RF, microwave and photonic semiconductors, has been granted a preliminary injunction in its lawsuit against Infineon Technologies Americas Corp over GaN-on-silicon technology rights.

The US District Court for the Central District of California in Los Angeles' decision confirmed Macom's continuing exclusive rights in certain GaN-on-silicon RF fields under a 2010 License Agreement entered into between Nitronex (acquired by Macom in 2014) and International Rectifier (acquired by Infineon in 2015); and it ruled that Macom is likely to succeed on its claim that Infineon's purported termination of that Agreement was improper and without effect.

It granted Macom's motion for a preliminary injunction prohibiting Infineon from engaging in activities inconsistent with the 2010 License Agreement pending the Court's final decision in the case.

"We were forced to file this lawsuit to stand up to Infineon's bullying and anticompetitive behaviour. We are gratified by the Court's preliminary decision confirming that the GaN-on-silicon rights granted to us under the 2010 License Agreement remain in full force and effect and that Infineon acted improperly in trying to operate in our exclusive field of use," said John Croteau, president and CEO of Macom.

"We are firmly committed to vigorously litigating this case to its rightful conclusion. We continue on the path to providing GaN-on-silicon technology that promises to improve network data service and cell coverage of 4G/LTE and 5G base stations that will benefit people worldwide."

The German Ministry of Economics has withdrawn its clearance certificate for the takeover the German semiconductor deposition firm Aixtron by Grand Chip Investment GmbH, a 100 percent indirect subsidiary of Fujian Grand Chip Investment Fund LP.

Fujian Grand Chip Investment Fund is a Chinese investment fund; 51 percent of which is held by the Chinese businessman Zhendong Liu and 49 percent by Xiamen Bohao Investment Ltd.

Back in May, the companies entered into a €670 million takeover agreement under which Aixtron shareholders would be offered €6.00 in cash per each ordinary share.

Aixtron has not given a reason for the reversal of the decision, but commentators in the press have highlighted growing concerns in Germany about the acquisition of cutting-edge technologies by Chinese firms and subsequent loss of knowhow and high-tech jobs.

Researchers at Dartmouth College, New Hampshire, working on a project called ‘DarkLight’ have developed and demonstrated for the first-time, how visible light from LED lighting can be used to transmit data even when the light appears dark or off.

The study, ‘The DarkLight Rises: Visible Light Communication in the Dark’ will be presented in New York at MobiCom 2016: The 22nd Annual International Conference on Mobile Computing and Networking on October 4th by Dartmouth co-author Zhao Tian, the lead PhD student for the project.

Through DarkLight, light-based communication is sustained even when LEDs emit extremely low luminance, by encoding data into ultra-short, imperceptible light pulses by using off-the-shelf, low-cost LEDs ($7 each) and photodiodes ($6-8 each).

The current DarkLight prototype supports 1.6kbps data rate at 1.8m distance.

"With DarkLight, we can potentially enable light sensing so that it is always on, 24/7, regardless of the light's illumination status," says the project's principal investigator Xia Zhou, assistant professor of computer science and co-director of the DartNets (Dartmouth Networking and Ubiquitous Systems Lab), which helped conduct the study. DartNets' research focuses on broad applications, systems, and networking perspectives of smartphones and smart device systems.

"DarkLight shows new possibilities on what visible light alone can do. We believe there are a lot more interesting applications yet to come," added Zhou.

In its 'Power GaN 2016: Epitaxy and Devices, Applications, and Technology Trends' report, Market research firm Yole Développement predicts that the GaN power business will reach $280 million in 2021, growing at 86 percent CAGR between 2015 and 2021.

It says the market is being driven by emerging applications including power supplies for datacentre and telecom, AC fast chargers, Lidar, envelope tracking, and wireless power.

"Numerous powerful developments and key collaborations have been announced during this period and confirmed a promising and fast-growing industry", comments Hong Lin from Yole. Collaborations include Integrated Device Technology (IDT) and Efficient Power Conversion (EPC); Infineon Technologies and Panasonic; Exagan and XFab; TSMC and GaN Systems for volume production, which all took place within the last two years.

Up until late 2014, 600V/650V GaN HEMTs' commercial availability was still questionable, despite announcements from various players. Fast-forward to 2016 (see above) and end users can buy low-voltage GaN (<200V) devices from EPC and high-voltage (600V/650V) components from several players, including Transphorm, GaN Systems, and Panasonic.

The idea of bringing GaN from the power semiconductor market to the much bigger analogue IC market is also now gaining interest, says Yole. For example, EPC Power and GaN Systems are both working on more integrated solutions.

LED company Cree has filed a complaint against E. Mishan & Sons, Inc. (Emson) with the US District Court for the District of Massachusetts for infringement of Cree's patented LED technology in Emson's flashlights, such as the Bell + Howell Taclight.

"Cree will not tolerate and will aggressively defend against such blatant disregard of its intellectual property rights," said Dave Emerson, senior vice president and general manager for Cree LEDs.

"We have an obligation to our customers to ensure that the market offers products that include the high-quality LED components that customers have come to expect. We are committed to protecting Cree's investment in research and development on behalf of our customers, shareholders and our licensing partners."

As part of the complaint, Cree is seeking an award of enhanced damages, attorneys' fees and an injunction to prevent Emson from offering for sale and selling any products using knock-off LEDs.

LayTec has delivered its 2000th in-situ metrology system since its foundation in 1999. An EpiTT with 2000 in its serial number has been shipped to Compound Semiconductor Centre (CSC) - a joint venture between compound semiconductor specialists IQE and Cardiff University.

CSC works on providing a complete capability value chain from high-end R&D through product and process innovation to high value, large-scale manufacturing. According to Wyn Meredith, director of CSC: "This EpiTT and other LayTec systems already installed in our labs provide unrivalled precision and sophisticated analysis algorithms, which is crucial for process optimisation in semiconductor manufacturing environment."

EpiTT - LayTec's workhorse for MOCVD mass production -  combines measurements of temperature and reflectance at three wavelengths in one tool. For True Temperature (TT), it usesEmissivity Corrected Pyrometry, which delivers the precise surface temperatures of opaque materials at 950 nm (Si, GaAs, InP). For materials that are transparent at 950 nm (GaN, Sapphire, SiC), EpiTT measures the temperature on the top side of the carrier. Measuring reflectance at three wavelengths monitors all essential properties of the growing layers, such as growth rate, film thickness, stoichiometry changes and morphology. 

LayTec's founder and CEO Thomas Zettler commented: "It is significant that our 2000th in-situ tool is delivered to a research institution with a strong connection to industry. LayTec has always set a great value on cooperating with both industry and R&D."

He added: “Until now, we have equipped hundreds of customers worldwide with state-of-the-art metrology, mainly in the field of LED and laser production. In the last few years we also entered the PV, display and advanced silicon markets. Meanwhile, our product portfolio covers all areas of process monitoring: in-situ, in-line, lab-line and map-line metrology. Due to this market diversification, we believe to deliver the next thousand tools much faster than before."

Thermal management specialist Cambridge Nanotherm has announced it has built significant additional capability and capacity into its MCPCB manufacturing base to meet rising demand for its award-winning thermal management solutions. The company has established partnerships with a wide network of PCB and thin-film manufacturers to offer a broad range of options in terms of circuitisation, quality, volume and standards.

Cambridge Nanotherm’s manufacturing capabilities include everything from fast turnaround prototyping, high-definition thin-film circuitisation, speciality manufacturing, through to high-volume mass production. Key industry and regulatory standards such as automotive standard ISO/TS 16949:2009, as well as industry-specific SGS standards, can be applied.

From LED chip packaging to high-brightness modules, thermal management is becoming a limiting factor as customer’s demand ever brighter LED devices in ever smaller footprints. To meet these requirements LED manufactures are being pushed into using more thermally effective substrates to ensure that LEDs stay cool enough to meet their advertised lifespan. Historically that meant switching from cost effective MCPCBs to expensive and difficult-to-work-with ceramics such as alumina and aluminium nitride. Cambridge Nanotherm offers an alternative. 

Sitting at the heart of high-power LED applications, Nanotherm LC and Nanotherm DM technologies are enabling a new generation of products that rely on effective thermal management to operate successfully. Cambridge Nanotherm’s proprietary LC and DM technologies are available exclusively via this manufacturing process.

Andy Matthews, COO at Cambridge Nanotherm, said: “Our MCPCBs offer designers a distinct thermal advantage. Demand for our solutions has therefore been strong, and we’re currently engaged with most of the top ten LED manufacturers. As a result, we’ve expanded our manufacturing routes to cater to a much broader variety of requirements. We’re working with some of the best PCB and thin-film circuitisation companies so we can offer an exceptional range of options to ensure we keep our customers satisfied. We will continue to develop our manufacturing capabilities to make sure we always offer the best available options to our customers.”

Nanotherm’s patented ECO process involves converting the surface of the aluminium core of the MCPCB, which acts as a heat spreader, into an electrically insulating but thermally conductive nanoceramic that offers outstanding thermal performance. Depending on the circuitisation route that is chosen, composite thermal performance of the resulting Nanotherm MCPCB ranges from 115 W/mK to 152 W/mK.

Standard LC products are covered by UL recognition, speeding up time to market for luminaires and modules. Nanotherm manages the entire process, from thermal design guidance and material choice to delivering the finished circuits. This makes the process seamless and simplifies the manufacturing route for customers.

Dialog’s delivery of GaN power ICs signals mainstream market adoption is close, reports Rebecca Pool. 

LATE THIS SUMMER, UK-based Dialog Semiconductor revealed plans to sample GaN power ICs in a fast charging power adapter by the end of this financial year. 

Following two years of collaboration with Taiwan Semiconductor Manufacturing Corporation, the power management IC supplier has developed a 650 V GaN power IC plus controller combination that is said to halve the size and power losses of silicon power management ICs.

“Until recently, GaN had been largely limited to small specialty fabs from vendors in Japan as well as research institutions,” says Mark Tyndall, Senior Vice President of Corporate Development and Strategy at Dialog. “In contrast, Dialog targets high volume consumer applications, so it’s never been our intention to prove GaN for, say, niche military applications.” 

“But when TSMC started to offer GaN as a standard process on six inch wafers, we saw that as a signal that this was the right time to enter this market, as well as solve the emerging problems of size in the power adapter market,” he adds.

Dialog’s end-result is the so-called SmartGaN DA8801; a monolithic IC that integrates enhanced-mode GaN-on-silicon HEMTs with analogue drivers and logic blocks in a 650 V half-bridge design for 25 W to 65 W adapters. By combining this IC with its ‘Rapid Charge’ power conversion controllers, Dialog can produce efficient, small, high-power-density adapters that look set to topple traditional silicon FET-based designs from pole position in the world of power adapters.

“We’re not trying to replace a silicon FET with a pure GaN transistor; how could a six inch [GaN-on-silicon] wafer process compete with [CMOS] manufacturer producing millions and millions of wafers?” points out Tyndall. “But we have optimised our solution around this GaN, half-bridge architecture which allows us to provide a GaN-based system at the same cost, or even lower, that the traditional silicon FET system.” 

“We’ve reduced the overall bill of materials in our systems and we’re getting a lot of excitement as we now have a small 45 W adapter in a 25 W adapter housing,” he adds.

Crucially, this reduction in size paves the way to the universal power adapter that the mobile communications industry craves for its smartphones, notebooks, iPads and more. And according to Tyndall, Dialog customers are more than ready for GaN.

While he reckons initial concerns centred on device reliability, times have changed and industry has moved on. “We’re really are beyond these issues now,” he asserts.

Indeed, as Richard Eden, senior analyst at IHS Markit, highlights, Dialog is one of many moving towards GaN. “It’s a safe bet that all silicon semiconductor suppliers are developing a strategy for GaN,” he points out.

“During the last five years, start-ups such as Efficient Power Conversion, Transphorm and GaN Systems have successfully developed GaN products for the commercial market,” he adds. “And more recently, larger silicon semiconductor companies, such as Infineon Technologies, Texas Instruments and Panasonic, have announced GaN developments.” 

For example, in March last year, Texas Instruments launched a 80 V GaN FET power-stage comprising a high-frequency driver and two GaN FETs in a half-bridge configuration.

For Eden, these developments and Dialog’s latest announcement are clear signals that GaN is now joining mainstream markets and is no longer considered to be a ‘clever research lab experiment’. “And with foundry companies like TSMC now offering a GaN-on-silicon transistor process technology to clients, perhaps we will see more suppliers offering similar products using their own driver IC [technology],” he says.

State-of-play

Right now, Dialog is testing its GaN-based power IC with ‘beta’ customers and is set to sample during Q4, this financial year. The company’s next move will be to ramp up production to volume levels, which Tyndall believes will take place by the middle of next year. 

“We’re strong in the fast-charging, mobile, power adapter market with around a 70 percent share and we’re also entrenched with the top ten vendors in the China smartphone market,” he says.

The Development and Strategy President doesn’t foresee any supply chain issues – manufacturing uses standard CMOS equipment – but highlights: “We do need to be careful with back-end packaging and have chosen experienced vendors here.”

“We also intend to use high volume, standard, device testers, which fits with our desire to manufacture hundreds of millions of units without using niche [processes],” he adds.

Then, with wireless device markets addressed, the company intends to turn to PC markets with a 100 W power adapter, and eventually target server markets.

Looking beyond high-volume consumer markets, Dialog isn’t chasing industrial applications such as PV inverters, turbines and electric vehicles, although according to Tyndall, the company would ‘never say never’.

However, the company has settled on GaN-on-silicon as its future technology of choice, over SiC, which Tyndall puts down to lower costs, better monolithic integration and higher volume production.“We are looking to drive production volumes now,” he concludes. “There is a chicken and egg element here as we need to drive volumes to drive the next level of fab investment that will bring us to a higher wafer size. But I would hope to move to eight inch wafers in the next two to three years.”

Richard Stevenson talks to IHS Markit analyst Richard Eden about the big events in the SiC industry, such as Infineon’s plans to purchase Wolfspeed, X-Fab’s establishing of a 150 mm SiC foundry, and the acquisition of Fairchild by ON Semiconductor. 

Q: Within the SiC industry, this year will be remembered for the buying of Wolfspeed, essentially Cree’s SiC division, by European electronics giant Infineon. As the first to bring the SiC diode to market, Infineon is already a leader in this technology. So what is the rationale behind its aim to buy Wolfspeed in a cash deal worth $850 million? 

A: There are two main reasons. Firstly, although Infineon is a big player in the SiC diode market, it’s way behind Cree and others when it comes to the potentially larger SiC MOSFET market. After several years of pushing its SiC JFETs, it only finally launched the SiC MOSFET earlier this year. So it’s four or five years behind the curve for SiC MOSFETs. By acquiring Wolfspeed, it will immediately get a product range that is established, which has gone through several design optimisations, and is already in mass production. So it can get market share. 

The other key element in the Wolfspeed acquisition is that Infineon obtains the world-leading substrate wafer production capacity, and its know-how. It gives Infineon in-house control of the essential resource to make its own wafers, and it saves costs from having to buy wafers from a commercial source. 

Q: Will this acquisition make Europe the leader of SiC power devices, given that STMicroelectronics also produces SiC diodes and MOSFETs based on this material? 

A: To some extent, yes. Control of Wolfspeed’s product development will shift across the Atlantic to Europe. But I don’t think that they’ll move the actual production or design engineering for a long time yet.

Q: How does the sale of Wolfspeed impact the US? Is General Electric now the local leader of this technology?

A: I think General Electric is probably one of the most well-known American companies, but there are plenty of other American companies that are active in the SiC market, maybe some smaller ones. GeneSiC, Global Power Technology Group, Microsemi, Northrop Grumman, Raytheon and United Silicon Carbide are all deeply involved, and you have new companies like Monolith Semiconductor. 

Q: In the Asia-Pacific region, is Rohm still the leading manufacturer of SiC power devices? 

A: Yes, I think so. But Rohm is unusual [in this region], in that it sells all of its SiC products commercially. It has no captive, internal market making end-equipment, whereas some of the other Japanese and Asian companies that produce SiC products do so for internal use in their finished end equipment. Selling [the SiC products] is a side line. For those companies, examples would be Denso, Fuji Electric, Hitachi, Mitsubishi Electric, Toshiba, Toyota and so on.

Q: ON Semiconductor has just bought Fairchild for $2.4 billion. Back in 2011, Fairchild equipped itself with SiC BJT technology by buying TranSiC. Do you think ON Semiconductor will continue to promote this technology? 

A: No, I don’t, because Fairchild stopped developing the SiC-based BJTs in around 2014. However, ON Semiconductor will continue to invest in wide bandgap materials, maintaining both companies’ strategies.  Fairchild had already announced SiC diodes, developed by the former TranSiC team in Sweden, with its SiC MOSFETs coming soon: this work will continue unchanged. ON Semi was working with GaN, in partnership with Transphorm, as well as an in-house GaN project: these will also continue. 

Q: Have all the world’s leading silicon power device makers now got SiC technology, either through acquisition or in-house development? 

A: I’d say no, but I would say that the ones who are not involved in SiC are probably working on GaN. The likes of Alpha & Omega Semiconductor, NXP, Panasonic and Texas Instruments are only involved in GaN.

Even Vishay – which has has historically bought up companies involved in older, established technologies – is developing a 650 V discrete, normally-off GaN power transistor in-house. This has not been formerly or officially announced, but this news was quietly displayed on the Vishay booth at Electronica. Expect plans to be announced in Q1 ‘17.  

Q: Do these electronics giants see SiC power devices as complementary to their silicon products, or as eventual successors?

A: In general, it’s a mix of both. Overall, they are hedging their bets, so they have an income stream in future, whichever way the market moves. In the short term it will be complementary, but in the long term it may be a successor – so they have got to be in that market, or their sales will just dry up. 

Q: Is there any appetite for the commercialisation of other SiC power devices, such as a thyristor? 

A:  Not that I’ve seen. 

Q: How will Infineon’s purchase of Wolfspeed impact the SiC substrate market?

A:  As Infineon will get full control of its SiC substrate manufacturing process, it can optimise the characteristic it wants for its products. It can allocate production to suit its needs as well.

I believe that Cree was the biggest supplier of wafers in the SiC power market, so Infineon may be able to control which of its competitors it supplies – if any. So I think the likes of Dow Corning, II-VI Incorporated and SiCrystal will be rubbing their hands, because the market-leading competitor may be about to exit the market.

Q: Dow Corning has strived to take a bigger share of the SiC substrate market. Has it made much ground so far, or does Wolfspeed currently dominate the substrate market? 

A:  I think Wolfspeed, or Cree, has still been the dominant supplier in the market up until now. I know Dow Corning has tried hard, and has caught Cree heavily, but I don’t think they have overtaken them. 

Q: Is the manufacture of SiC chips shifting to larger substrates. 

A:  Yes, but slowly. The market is still using 100 mm diameter wafers at the moment, but it is actively transitioning to 150 mm wafers. In my most recent report, published in February this year, I estimated the use of 150 mm wafers would overtake 100 mm by 2018. 

Q: Are there still issues over substrate quality, such as micropipes and Basal plane dislocations, or are these now a thing of the past?

A:  I think they are gradually being solved, but they are still a problem for the lower price wafer suppliers. So, without wanting to point the finger, there are several Chinese suppliers that have come on to the market with very cheap quotes. These [imperfections] are real problems for them, compared to the high-quality, more Western suppliers. You get what you pay for. 

Q: What is the SiC market worth today? And what proportion of it is for diodes and for transistors? 

A:  In my most recent report I estimated that the total SiC power market was worth about $200 million in 2015. Of that, SiC diodes comprised about 60 percent, SiC transistors just under 20 percent – and that was dominated by SiC MOSFETs – and SiC power modules comprised just over 20 percent. That’s mainly in hybrid SiC modules, so modules which combine silicon transistors with SiC diodes.

Q: What is the typical operating voltage of these products? 

A:  It depends on the device type. At the moment, most SiC diodes are 650 V rated, most SiC MOSFETs are 1200 V, and power modules are split roughly 50-50 between 1200 V and 1700 V.

Q: Many have argued that GaN HEMTs are suited to providing operation below 1 kV, and probably below 600 V. But recent research has shown that GaN HEMTs can operate well above 1 kV. So is the SiC power device market under a significant threat from GaN?

A:  Maybe, but it will take a very long time. It will take several more years for GaN transistors to become very common and successful commercially, and it will take a lot longer to get over 1000 V GaN to be perfected – ten to twenty years, at least.

Q: Where are the majority of SiC power devices deployed today, and how might this change over the remainder of the decade?

A:  In 2015, the biggest application was AC-DC power supply units, mainly used in computing or telecommunications. By 2019, I think the largest application will be hybrid and electric vehicles. We are already seeing SiC diodes used inside the DC-DC converters on hybrid vehicles. They have been in production for about one to one-and-a-half years. Hybrid vehicles will become much more common, and the use of SiC in the vehicles will increase too. So you have two growth factors.

BY JIANG LI, TOM BROWN, MEHRAN JANANI, JIRO YOTA, AND CRISTIAN CISMARU FROM SKYWORKS SOLUTIONS

Smartphones are delivering staggering levels of performance. These marvelous devices can stream videos with great resolution, provide music on the move, take great pictures, and run a vast number of applications.

To accomplish all of this and more, numerous components are crammed together on a very small footprint. So there is great pressure to reduce the real estate of every part, including the power amplifier that enables communication between the handset and the base station.

The dominant transistor technology for the power amplifier is the InGaP/GaAs HBT. It delivers excellent RF performance in a reproducible manner, and it is manufactured in high volumes with a high yield. However, when this chip is deployed in advanced frond-end modules, it fails to meet RF designer’s zero-size requirements. Given this state of affairs, there is a compelling need to integrate the HBT with a FET or a pHEMT on the same chip, because this aids design flexibility and circuit functionality while reducing the module’s size and its overall cost.

One option for forming a single GaAs chip that integrates either the FET or the pHEMT with the HBT is to insert the FET into the emitter section of the HBT. Compared to a HBT-only process, the strengths of the resulting BiFET are a low epitaxial cost, shared process steps, and minimal impact on production cycle time. But there are also weaknesses – as the FET and HBT have shared layers, optimization of the device is hampered; and the BiFET has poor RF isolation.

At Skyworks Solutions we have pursued the second of the two options for device integration, successfully developing an advanced BiHEMT technology that stacks the pHEMT underneath the InGaP/GaAs HBT sub-collector. To enable commercial success, we have identified and addressed many fabrication challenges. This has led to high volume, high-yield manufacture of GaAs BiHEMTs on our 6-inch production line.

Forming BiHEMT epiwafers is straightforward, involving the growth of the HBT epitaxial structure on top of the pHEMT. Using MOCVD, HEMT epitaxial layers are deposited on 150 mm semi-isolating GaAs substrates, before the highly doped HBT sub-collector is added, followed by the collector, base, and emitter layers.

We have introduced slight modifications to our HBT layer stack to accommodate the implementation of the pHEMT underneath the HBT. This includes the optimization of the HBT sub-collector thickness, in order to reduce the overall BiHEMT topology. The electric performance of the HBT is not compromised by this adjustment.

To fabricate the BiHEMT, we use many conventional processes (details are provided in the box “Making a BiHEMT”). Compared to the standard HBT-only process, the additional process challenges for making the BiHEMT are primarily associated with the severe topology created by the combination of active devices, passive components, metal interconnections, and dielectric layers. There is a step height of about 2 mm between the top of the HBT emitter contact and the pHEMT gate layer at the bottom (see Figure 1).

Topography challenges

Due to this substantial height difference, it is critical to optimize the photo and etch processes for gate formation, so that the pHEMT gate Schottky layer is opened and etched with good uniformity and process control. These processes need to be reliable and provide a good process margin, to prevent wafer scrap that leads to increased loss and inferior
yield.

Preventing shorts

One weakness with this approach is that the photo resist is not perfectly conformal – it is far thinner on the top of the emitter of the HBT than it is at the TaN level (see Figure 4 (a)). Note that on top of the HBT emitter, the resist can be less than 100 nm-thick. As a result, during SiN dry etching for the TaN resistor, the photoresist on top of the HBT emitter can be fully eroded, resulting in TaN metal deposition on top of the HBT emitter. Ultimately, this can prevent a connection between the HBT Metal-1 metal and the emitter metal; and it can result in an emitter short and yield loss (see Figure 4(b)). This yield loss is 3 percent at the final probe test.

A better way is to combine a moderate increase in resist thickness with the selection of a more conformal photo resist, to improve resist coverage. By taking this approach, we have tackled yield loss without impacting electrical performance.

Additional issues that have to be addressed when working with the severe BiHEMT topology occur in backend-of-line process modules. There is the possibility of inter-level metal interconnection shorts and/or poor isolation (an example of this is shown in Figure 5, which reveals shorts between Metal-2 and Metal-3). This type of problem is caused by the severe underlying topology, and is associated with poor material coverage of the photo resist and inter-level dielectric. When the inter-level dielectric is dry etched, the resist – and even the dielectric material – are removed, resulting in shorts between the interconnection metals. Yield loss is exacerbated by the presence of metal stringers. This form of metal short is fatal, and can result in wafer scrap. 

Making a BiHEMT

Production of the BiHEMT begins with the formation of the emitter contact. Photolithography defines the HBT emitter pattern, and emitter contact metal evaporation and lift-off follow. There is no need to employ an emitter alloy, because emitter metal contacts are deposited on a highly doped emitter top layer.

A combination of photolithography and an inductively coupled plasma dry etch, which stops on top of the InGaP HBT emitter layer, defines the emitter mesa. This pair of processes is also used to form the base pedestal. Deposition of a SiN film by PECVD provides protection to the HBT active area, while access to the pHEMT layers comes from wet-etching of the sub-collector layer.

Formation of the HBT base contact involves etching of the SiN film and the InGaP layer, metal deposition and liftoff; the HBT collector contact is defined by deposition of AuGeNi/Au and subsequent rapid thermal processing. The pHEMT source and drain ohmic contacts are formed simultaneously with the HBT collector. Devices, diodes, and other passive components are electrically isolated by helium ion implantation, and the pHEMT gate defined by photolithography, gate recess wet etch, Ti/Pt/Au gate metal deposition and lift-off. 

Following formation of the intrinsic HBT and the pHEMT, devices are passivated by PECVD of SiN films. The next steps involve the fabrication of TaN thin film resistors, metal-to-metal SiN capacitors, inductors and metal interconnects. The front side process for the BiHEMT is completed with final passivation, resulting from SiN deposition.

BY JYH-CHIA CHEN, MILTON YEH AND HUSSEIN S. EL-GHOROURY FROM OSTENDO TECHNOLOGY INC.

There are fundamental flaws with both of the leading approaches for producing LED-based white light sources and full-colour displays. If white light is formed from the colour mixing of separate red, green and blue LEDs that are packaged together as one device, the resulting unit must combine red LEDs made from AlGaInP with GaN-based green and blue emitters – and pairing these different types of LED requires special efforts, due to substantial differences in electrical characteristics and material properties. With the main alternative, there are also major downsides. In this case, a blue LED is coated with a yellow phosphor, and white light results from colour mixing. Efficiency is then compromised, due to cascade losses that occur during the blue-to-yellow colour conversion of the photons.

In addition to these drawbacks, both of these approaches are held back by basic issues related to the source, including a high packaging cost, low efficiency, a large device size, low yield and poor reliability. What’s needed, but not commercially available, is a single device that emits all the three primary colours of light.

Addressing this limitation is our team from Ostendo Technology Inc. of Carlsbad, CA, which has created the world’s first LED that delivers controllable, tuneable full-colour emission through adjustments in current flow. Note that in addition to the three primary colours, our device produces other colours through colour-mixing techniques.

An incredibly important aspect of our technology is that it involves a growth process that is identical to that used for commercial, single-color LEDs – there is no additional pre- or post-growth process. Thanks to this, our technology is practical, and can be scaled to high-volumes and low costs.

Full-colour design

Unlike conventional III-nitride LEDs, we use a different quantum well section for each primary colour. Between these sections are specially-designed intermediate carrier blocking layers, made from AlGaN layers with different thicknesses and dopant concentrations (see Figure 1 for a cross-sectional view of our monolithic, three-colour InGaN-based LED).  

We have found that the intermediate carrier blocking layers play a key role in controlling the transport and distribution of the carriers across the active region. Careful selection of the thickness, composition and dopant concentration of each of the intermediate carrier blocking layers enables control of the transport of electrons and holes under various current injection conditions. Thanks to this, electrons and holes can be directed into quantum wells that lead to emission at specific dominate wavelengths. 

Our monolithic LEDs are grown by MOCVD, using a close-coupled shower-head reactor, on c-plane (0001) sapphire substrates. They comprise: a 2 mm-thick, undoped GaN layer; a 3 mm-thick, highly silicon-doped n-type GaN layer; an InGaN compliance layer, inserted for quality improvement purposes; blue InGaN/GaN multi-quantum wells emitting at 460 nm, which contain three 3 nm-thick In0.15Ga0.85N wells separated by 10 nm-thick InGaN barriers; a green-emitting section producing 530 nm emission from three 3 nm-thick In0.25Ga0.75N quantum wells separated by 10 nm-thick InGaN barriers; a region producing red, 650 nm emission from a single 3 nm-thick In0.35Ga0.65N well sandwiched between 10 nm-thick GaN barriers; a 20 nm-thick, magnesium-doped p-type Al0.15Ga0.85N electron blocking layer; and a 200 nm-thick p-GaN contact layer. 

In addition, we have investigated electroluminescence, collecting data from multiple positions on each epiwafer. This has been undertaken with an in-house ‘Quick Tester’ attached to an integrating sphere. Using indium-ball contacts, typically 0.8 mm in diameter, we have obtained normalized electroluminescence spectra for devices under different injection currents (see Figure 3). In contrast to the photoluminescence results, only one predominant emission peak appears at one particular current. As the injection current is cranked up from 15 mA to 200 mA and then on to 400 mA, emission shifts from red (~650 nm) to green (~530 nm) and finally blue (~460 nm). 

Ultra-high-resolution displays

Our technology is of great interest to LED and display manufacturers, because it can create micro-LEDs that produce: a far higher luminous efficiency than organic LEDs; and thanks to their self-illumination, eliminate backlight demands associated with LCDs. Displays made from microLEDs are also more robust than those incorporating organic LED and LCD technologies, making them better-suited to harsh outdoor applications. And micro-LEDs can be housed on flexible substrates, making them excellent candidates for wearable devices.

One group that has made significant strides with displays employing microLED arrays is that of Kei May Lau and co-workers from Hong Kong University of Science and Technology. This team recently demonstrated a micro-display with a resolution of 1700 pixels-per-inch that features micro-LED arrays integrated on silicon CMOS substrates. This micro-display has the potential to serve many applications, such as active driving of pico-projectors, augmented reality, head-up displays in cars, and headsets for gaming. However, the work by this group has been limited to blue light of differing intensity. Turning to our LEDs, and using them in combination with the bonding-on-silicon technology, would enable full-colour displays, without adding any complexity to device fabrication.

In short, we believe that our novel, InGaN-based LED technology that incorporates intermediate blocking layers will deliver a significant reduction in the fabrication complexity of full-colour micro-displays. In addition, it will allow large-scale flexible displays to be realized that can compete favourably with existing organic LED technology. We are now developing our technology for displays, which will also be a game changer for next-generation lighting, due to its efficiency, compactness, simplicity and robustness.

Green LEDs formed from cubic GaN can have lower Auger losses, higher hole carrier mobility, greater optical gain and increased p-doping efficiency

BY CAN BAYRAM FROM THE UNIVERSITY OF ILLINOIS AT URBANA-CHAMPAIGN

The revolution in solid-state lighting is well underway. LEDs bulbs are now shipping in significant volumes, thanks to price tags that are far more competitive than they were just a few years ago.

However, if sales of this class of lighting are to continue to increase, it must not only get cheaper – it must also deliver higher powers and a better colour quality. And that means a fundamental change in the emission process. 

All of these six issues plaguing these conventional green LEDs, which employ the hexagonal phase of GaN, have been studied in significant depth. It is well known that the high indium content needed to ensure that InGaN quantum wells emit in the green LED hampers the efficiency of the device, due to a miscibility gap in this alloy; and that the epilayers are riddled with defects, due to lattice- and thermal-mismatch, which arise from growth on foreign substrates, such as sapphire and silicon. In addition, realising a high level of p-doping in nitrides is challenging, due to a high activation energy for the magnesium impurity (it is in excess of 200 meV); there is also a significant asymmetry between electron and hole densities and mobilities, stemming from the large difference between activation energies for p-type and n-type dopants; there are strong internal electric fields within the LED, associated with non-centrosymmetry and inherit piezoelectricity of the hexagonal lattice; and last, but by no means least, Auger non-radiative recombination supresses efficiency at current densities beyond just 10 A cm^-2.

Making matters worse, the high-quality epitaxial regrowth of GaN on non-polar surfaces is challenging, due to the low surface energy of non-polar planes; and the quality of InGaN alloys is inferior on this plane. So, in short, the m-plane platform is far from ideal for enabling polarization-free LEDs – and lasers – to fulfil their promise.

A better option for fabricating polarization-free GaN devices is the cubic phase of GaN (see Figure 1). Structures with this phase are completely free of spontaneous and piezoelectric polarization along the common <001> direction. What’s more, the optical gain of quantum wells from the cubic phase of GaN/InGaN exceeds that of hexagonal variants, thanks to the combination of lower effective masses for the electron and hole; and weaker polarization in the transverse-electric direction, due to a small spin-orbit splitting energy.

An alternative approach is to incorporate an impurity, such as manganese, into the growth. This transforms the hexagonal phase of GaN into a cubic one. However, as this technique relies on p-d orbital repulsion between the 3d impurity levels, success hinges on realising a high density of impurity incorporation. That’s highly undesirable, because it degrades material quality to such an extent that it is unsuitable for device fabrication. 

Groovy nitrides

We are exploring another avenue, forming cubic III-nitrides in U-shaped [silicon{111}- silicon{100}- silicon{111}] grooves. Our success hinges on the equivalence of the h-crystal <0001> direction and the c-crystal <111> direction. This property means that when two h-phase growth fronts merge at ~110° – that is the angle between the two Ga-N bonds in the hexagonal tetrahedral bonding – a c-phase forms after the seam (see Figure 3).

There are several ‘levers’ for optimising the transition between the hexagonal and cubic phases. Elimination of hexagonal incursions and minimizing defectivity in cubic phase material results from perfecting the opening width, oxide thickness and etch depth.

Our work lays the foundation for cutting droop in green LEDs. The biggest contributors to this malady in conventional devices are high indium content, material defectivity, low p-doping, asymmetry, polarization, and Auger recombination. Switching to a cubic phase allow many of these weaknesses to be tackled head on – it enables smaller effective masses, higher drift velocity, higher carrier mobility, higher doping efficiency, higher optical gain, and smaller Auger losses. So it is clear that a promising route to efficient green LEDs – and ultimately efficient, high-quality LED lighting – is to use the cubic form of this material.

This work was carried out in the Micro and Nanotechnology Laboratory and Frederick Seitz Materials Research Laboratory Central Research Facilities, University of Illinois at Urbana-Champaign, IL, USA.