High-quality growth of LEDs on silicon is notoriously difficult, with stresses and strains causing wafers to bow. But these problems are not insurmountable: Plessey Semiconductors is now churning out GaN-on-silicon LED chips from flat epiwafers using a recipe developed at the University of Cambridge. Richard Stevenson reports.

 
 
You’ll need deep pockets to buy an LED bulb that can replace a 60 W incandescent. If you insist on the best that money can buy, you’ll go in search of the Philips bulb that won the Department of Energy’s Lighting Prize, and have to fork out $30.

If you are willing to sacrifice efficiency and colour quality, you’ll be able to trim a few dollars off your outlay, but you must still expect to pay over $20.
 
Retailing for these prices makes the 60 W-equivalent LED bulb the preserve of the early adopter. Claims of 15 year lifetimes and efficiencies exceeding those of the compact fluorescent are undeniably attractive, but most people are not going to get their wallet out until prices fall substantially – maybe to $10 or less.

 
An X-ray diffraction tool scrutinizes the crystal quality of the LED epiwafers.

 Bulb prices will tumble when LED chips are made more cheaply, because these devices account for a very high proportion of the cost of the bulb. According to the most recent Solid-State Lighting Manufacturing Roadmap that was published in August 2012, in 2011 the LED package accounted for more than half of the total manufacturing cost of the bulb.
 
One of the most promising options for trimming the costs of LED manufacture is to switch substrates, replacing sapphire with silicon. This platform is not just cheaper; it allows wafer processing on lines that were installed many years ago for making silicon-based products, so there is the opportunity to either create LED production lines with minimal capital expenditure, or to outsource LED manufacture at very competitive rates.
 
Savings associated with making LEDs on silicon are well known within the industry, with several companies trying to develop technologies for producing this type of device. Lattice Power from China has arguably led the way, bringing to market the first white-emitting, GaN-on-silicon LEDs, which were made on 2-inch GaN substrates. It is developing a process for 6-inch silicon wafers, which should enter production this year.
 
Meanwhile, the Californian outfit Bridgelux has developed warm-white LEDs in it labs that deliver 125 lm/W at a drive current of 350 mA, and its technology is now being used by Toshiba to make devices on 200 mm silicon. In addition, the likes of Samsung, Osram, Azzurro and Plessey Semiconductors are developing technologies for GaN-on-silicon LEDs on 6-inch or 200 mm silicon substrates.
 
Although Plessey has no background in LED manufacture, it could well be the pacemaker in this sector: Its management is adamant that it has a very competitive technology.
 
“We’re convinced we are 18 months ahead of anyone else that is talking about GaN-on-silicon 6-inch,” says company chief operating officer, Barry Dennington. He has found it “impossible” to acquire any 6-inch epiwafers, and concludes that rivals firms are yet to produce a sustainable source of this material.
 
Plessey has been sharing its GaN-on-silicon LEDs that are fabricated at its Plymouth fab with potential customers since autumn 2012, and the company plans to launch its first products in the first quarter of this year.

 



Plessey is manufacturing its GaN-on-silicon LEDs on an Aixtron reactor with a capacity of seven 6-inch wafers. The tool is fitted with various in-situ monitoring tools, which measure the curvature of the wafer and its temperature.

The GaN-on-silicon technology that Plessey has originates from the research group of Sir Colin Humphreys at the University of Cambridge. The GaN-on-silicon growth process developed there formed the core IP of the spin-off CamGaN, which Plessey acquired in February 2012.
 
This IP addressed the biggest challenge of producing GaN-on-silicon wafers, which is the deformation of their geometry that stems from substantial differences in the expansion coefficients of GaN and the underlying substrate. Wafers that are flat at the deposition temperature, typically 1000 °C, can bend and flex during cooling, so that when they are removed from the growth temperature, they can be bowed to such an extent that the epiwafer is cracked and cannot yield working devices.
 
It is possible to prevent this from happening by inserting a carefully selected stack of layers into the epiwafer that address the stresses and strains in this structure. The researchers in Humphreys’ group have mastered this on their single-wafer 6-inch reactor, using the combination of an AlN nucleation layer, a complex buffer structure and layers of AlGaN and GaN. Inserting a SiN layer into this structure cuts the threading dislocation density.
 
Dennington claims that one of the great features of the Cambridge recipe is the complete neutralisation of the bow and mismatch in the crystalline structure between silicon and GaN. The result is particularly impressive because these wafers, which are sufficiently flat for processing in silicon lines, have a buffer that is just 2.5 µm thick. “We think competition is around 6-8 µm-thick,” remarks Dennington. “So ours is thinner, helping us with wafer bow, and giving us better throughput through the reactor.”
 
Although the acquisition of CamGaN by Plessey is a perfect fit, it is not one that many would have foreseen during the latter part of the previous decade. It’s not just that the Plessey 6-inch silicon line was in the hands of X-fab at that time (see box “The evolution of Plessey Semiconductors”); back then Humphreys’ team was involved in a £3 million GaN-on-silicon LED project headed by RFMD’s Newton Aycliffe operation, and involving QinetiQ, Forge Europa and the UK branch of Aixtron. It seemed that if LEDs were ever going to be made in high volume in the UK, production would be out of the RFMD fab.
 
However, although the engineers at the Plymouth site were not involved in this project, they have been more than just observers in the development of this technology. “[Humphreys] had grown some GaN-on-silicon wafers, and we had a 6-inch wafer facility, so we took some wafers and helped with some very early LED structures,” explains Dennington. Plessey Semiconductors’ CEO, Micheal LeGoff, kept a close eye on the progress of this technology, and then snapped up the company in February 2011.
 
“We don’t know who else we were bidding against,” remarks Dennington, implying that he doesn’t know if Plessey fought off bids from the likes of RFMD. “Obviously it’s a business deal, but I do believe there was some passion amongst some of the people involved in CamGaN and Cambridge to see that this technology would be planted in a British company.”
 
Plessey’s acquisition also led to its participation in a European Consortium called “Consumerising Solid-State Lighting”, which had previously included Cambridge University and QinetiQ as partners in the project. “The Consumerising Solid-State Lighting programme was chaired by Philips, and the focus was to create a $9.95 lamp that would replace the 60 W incandescent bulb,” explains Dennington, who adds that Plessey is the only LED manufacturer in that programme.
 
Alongside the transfer of IP to the Plymouth fab, the acquisition of CamGaN involved the transfer of a handful of employees, who are former post-docs from Cambridge University. “We have plans in the future to open up a Cambridge lab, so that our Cambridge employees can work locally,” says Dennington, who explains that these staff currently split their time between the fab and offices within the university.
 
After the acquisition of CamGaN closed, the former post-docs had access for one week every month to the Thomas Swan single-wafer reactor, which they used to produce 6-inch GaN-on-silicon material. But since the arrival at the Plessey site of the Aixtron Crius II XL, a reactor capable of accommodating seven 6-inch wafers in a single growth run, they have focused their attention on the higher throughput tool.
 
Aixtron or Veeco?
 
The Plessey management found it tough to choose between an Aixtron and a Veeco tool for their 6-inch manufacturing line. “They are both excellent machines,” says Dennington, “and both companies were very, very eager to be associated with Plessey.” The development of the growth technology on an Aixtron showerhead machine tipped the balance in favour of a tool from the Aachen outfit.
 
Getting the Crius II XL up and running has gone very well. “[It] was delivered, installed and commissioned on time,” reveals Dennington. Since then, engineers at this site, supported by technical staff from Aixtron, have employed a step-by-step approach to developing the 6-inch GaN-on-silicon growth process, which began with GaN deposition on 2-inch sapphire. The quality of the material produced with this process met the expectations of the Aixtron staff, who use this as a benchmark for installation. 6-inch sapphire then replaced the 2-inch substrates, prior to the introduction of the Cambridge recipe and standard silicon wafers with thicknesses of either 750 µm or 1 mm.
 
One of the features of the Thomas Swan reactor installed in Humphreys’ group is its comprehensive suite of in-situ monitoring tools: It has a Laytec instrument for determining wafer bow from measurements of light reflectance off of the epiwafer’s surface, and it is equipped with an Aixtron Argus instrument for determining temperature profiles. Both tools are fitted on the Crius XL II to monitor growth processes within the reactor.
 
The acquisition of CamGaN did not give Plessey a finished product: It just received a process capable of repeatedly growing, on silicon, flat nitride epiwafers with active regions producing internal quantum efficiencies of 80-85 percent. The engineers at Plymouth have taken this as a basis for producing a range of LEDs.
 
“We’ve been able to produce lateral LEDs for some time – there’s no challenge in that – and we are now focusing on the vertical LED,” explains Dennington. Light extraction in this class of LED is enhanced through the introduction of mirror layers and a roughening of the chip’s surface.
 
Modifications to the 6-inch line to equip it for LED manufacture have been relatively minor, and include the addition of a wafer bonder and an X-ray diffraction tool for scrutinizing epilayer crystal quality. Operating at three growth runs per day, with seven 6-inch wafers per load, the line is capable of churning out 2 million 1 mm by 1 mm LEDs every week. And this capacity is set to increase, because Plessey’s business plan includes the installation of three more MOCVD tools. “With the efficiency of running four machines together, we don’t think we just have to multiply 2 million die by four,” claims Dennington. “We have modeled that we can get up to 10 million die per week from four reactors.

”Once the dies are made, they are sent off-site for packaging, as either single die or chip-board products. “We have an external partner,” explains Dennington, “which is one the traditional IC assembly houses that wants to move into this area.”
 
It is also possible that Plessey will team up with a strategic partner that will make GaN-on-silicon LEDs overseas. “This is partly likely because of capacity,” explains Dennington, “but as we are more successful at selling higher volumes to certain customers, they will expect a dual source through risk management.”
 
To accommodate those wishes, the management at Plessey have started discussions with Asian lighting companies.
 
The company has wasted little time in using the technology it acquired to make LED products. In autumn 2012, it provided potential customers with ‘demonstrators’, which are LEDs that indicate the latest progress in device performance, and it provided quotes, some for sales to 2015. And early this year it is releasing 0.4 mm by 0. 4 mm and 1 mm by 1 mm LEDs.
 
Thanks to the low cost associated with the GaN-on-silicon platform, one would expect that LED prices can be competitive, while still generating a healthy bottom line for the company.
 
As Plessey’s sales rise, it’s possible that other GaN-on-silicon LED manufacturers will also grow their revenues. This means that there could then come a time when claims of patent infringement are rife, as companies try to gain the upper hand over their rivals.
 
“In the LED market, acquisitions are made to capture IP: It’s quite a weapon,” argues Dennington. “But we are confident that we have got a very powerful IP.” The company is currently filing patents. However, these do not detail some of the black-art related to epitaxy, which Dennington and his colleagues believe will remain a secret even if rivals get their hands on Plessey’s LEDs and probe them with various techniques.
 
The company’s vision for its future goes beyond being a tremendously successful GaN LED manufacturer, and includes a move into the production of smart lamps. “We could expand the fab here,” enthuses Dennington. “We have a space at the back of this building to build another factory. It could be wafer manufacturing, or it could be a lighting company.”
 
The latter move could be a smart one, because it could ensure success after LED light bulb sales have plateaued and homes are full of affordable, long-lasting bulbs drawing just a handful of watts, but putting out the equivalent of a 60 W incandescent.
 
© 2013 Angel Business Communications.
Permission required.



Silicon absorbs the light produced by the LED, so it is removed with a process that begins with bonding the epiwafer to a carrier wafer. 



The evolution of Plessey Semiconductors 

UK-based international electronics, defence and telecommunications company Plessey formed in 1917, and forty years later it created a division known as Plessey Semiconductor. This off-shoot initially produced devices from fabs based around Swindon, but in 1987 it boosted capacity with a site in Plymouth that was opened by Prince Charles.


The Plymouth fab featured a 6-inch silicon line that was built to fulfil a large demand for telecom products, which were sold to Plessey Telecommunications and other telecoms companies. However, in addition to these products, which were deployed in an emerging digital telephone infrastructure, the company made components for the military, the automotive industry, and for TVs and set-top boxes. In addition, Plessey Semiconductors manufactured products for GSM basestations, winning contracts with Ericson mobile.

In the late 1980s, Plessey acquired Ferranti Semiconductors, a move that provided the firm with additional manufacturing facilities in Oldham, Manchester. And in 1995 Plessey was involved in another acquisition, with GEC taking over the Portsmouth site that was subsequently known as GEC Plessey Semiconductors. This takeover brought further investment, with the installation of an 8-inch line. However, at that point GEC owned Marconi Microelectronic Devices, and it had more manufacturing sites around the UK than it needed.

Plessey Semiconductors was put up for sale and bought by the Canadian telecoms giant Mitel. The firm split in two, retaining its name for its telecoms business, and launching Zarlink to represent its semiconductor efforts.

Ownership of the site by Zarlink didn’t last long, and in 2002 X-fab took it over, running the plant as a foundry. By 2009 this facility did not fit with the needs of that company, and in 2009 Michael LeGoff, Plessey Semiconductor’s current CEO, acquired it, plus the Swindon fab that was owned by Zarlink.