Playing catch-up
Indium phosphide wafer fab processes have traditionally lagged several years behind those used for silicon and gallium arsenide. But now the sector is beginning to catch-up, thanks to growing use of automation and better process control. Mike Syrett, director of Bookham's Caswell wafer fab operation, reports.
Indium phosphide (InP) supports a vital and maturing technology, and is used, for example, in many optical communications devices, both for the telecom and military markets. There are many reasons for InP's importance in these sectors. Typically InP gives a very compact, lower-cost modulator technology for 2.5Gbit/s and medium-reach systems at 10Gbit/s speeds. InP also allows the integration of multiple functions - such as lasers, modulators and variable optical attenuators - on a single chip or in a single package. This matters because customers - whether system vendors, telecom operators or the military - continually demand smaller, cheaper, more scalable and more functional devices and subsystems.
And they want them now - or at least not in two years' time. This is one area where InP fab capability has historically had a problem. InP technology has traditionally operated on 50mm-wafer fab processes, relying largely on tools adapted from the silicon industry 20 years ago - tools that are no longer supported by the semiconductor equipment manufacturers. Silicon (Si) and gallium arsenide (GaAs), the technologies that dictate the direction of this industry, have moved on in terms of wafer size as well as automation and process control.
So how can InP fabs meet the key challenges of lower costs, improved device performance, faster time to market and higher functionality through next-generation integration? It is reasonable that there should be advantages in moving from the essentially manual 50mm-wafer processes to the more sophisticated and automated larger wafer processes and tools long used in the Si industry, which, due to its market size, is well supported by the semiconductor equipment industry. The questions are whether the move can be done and whether the potential advantages can be realised.
Bookham has proved that this can be achieved (see panel "Gearing up Caswell InP fab for 75mm-wafer technology"), and now operates what the company believes is the world's only 75mm-wafer InP fab for telecom products. In doing so Bookham has put into practice a number of improvements from the Si and GaAs semiconductor industry.
Bookham achieved this capability through its acquisition of Nortel Networks' Optical Components businesses (NNOC), which launched 75mm-wafer InP in the late-1990s to boost capacity in the then booming 10Gbit/sec optical carrier markets.
Experience in operating the new 75mm-wafer has shown that greater automation and process control from advanced semiconductor tools reduce costs significantly. There are obvious cost savings in replacing manual processes with faster and more reproducible automated processes. A particularly important gain is the improvement in yields. This occurs at several levels. Fundamentally, the process uniformity is much better across a single wafer and also from wafer to wafer. At a higher level, however, it is now easier to push yield issues much further back into the earlier stages of the manufacturing flow. This is due to a combination of improved control and reproducibility, and the use of in-line measurement techniques and wafer mapping.
Wafer mapping creates 'maps' of certain measured parameters across a wafer as it passes through the processing steps. This allows operators to spot visible patterns and correlations in various electrical and optical characteristics, helping them to pinpoint problems in the fab processes. Adjustments can then be made to improve the yields of end devices. The more stable, reproducible and controllable the processes are, the easier the correlation and adjustment.
Cost saving is also made from lower materials wastage, as 75mm wafers give proportionately more die per wafer than 50mm wafers. Importantly given the current state of the market, the 75mm-wafer process gives good cost savings even when running at relatively low capacity levels. This is partly due to the improvement in wafer yields, but also because there is a significant improvement in overall costs from the better measurement and testing during processing, as this reduces the number of packaged devices that are faulty.
As important as cost savings are, the 75mm-wafer technology goes much further, and devices can be manufactured with a higher performance than is possible with 50mm-wafer processing tools. Mach-Zehnder modulators, for example, depend critically for their high performance on an extremely accurate geometry for the junction of their two waveguide arms. The resolution and accuracy required in photolithography, for example, is achieved with state-of-the-art production stepper lithography tools and is simply not viable using production 50mm-wafer contact lithography processes.
Crucially, Bookham has found that time to market is much faster with the new process, which gives faster cycle times for deposition, etching and so on, achieved through automated single-chamber load-locked cassette-to-cassette tools. It allows new products to be developed straight into the manufacturing fab, cutting out the costs and delays of using a preliminary R&D fab and then transferring to the manufacturing fab. It is also possible to use multiple and remote design teams, as is common for Si and GaAs fabs. Bookham has design teams at its Caswell, UK and Ottawa, Canada, sites.
Learning is also naturally quicker in this situation, particularly as existing processes can be used at the design stage where appropriate. Bookham is able to use a combination of design-of-experiments and device modelling to increase the proportion of designs that work first time.
The end result is that the 75mm-wafer process gives a highly scalable operation. This means that once a company has demonstrated a working device it can be confident of reproducing it and ramping up volumes to meet demand.
Box1: InP, Si and GaAs
Some of the processes used for InP are the same as for Si and GaAs - for example, photolithography and dielectric deposition and etching. But there are significant differences: in the front-end stages, the epitaxy process, while having some similarity to that of GaAs, is more critical and harder to achieve technically; there are also significant differences in the back-end processes, and these have big impacts on costs. Crucially, InP lasers are edge- not face-emitting. This means that their optical characteristics cannot be measured in the wafer state. Instead, the wafer has to be broken into bars and the bars coated on their front and back edges before optical measurements can begin - and ultimately the bars themselves have to be cut into separate chips to complete the measurements. So testing InP devices before they are assembled into a final packaged product is challenging but essential if rejects at the packaged stage are to be minimised.
Box2: Gearing up Caswell InP fab for 75mm-wafer technology
Bookham acquired Nortel Networks Optical Components (NNOC) in November 2002 and immediately began the transfer of NNOC's 75mm-InP-wafer fab from Ottawa, Canada, to its site in Caswell, UK. Bookham validated a new Class-100 cleanroom in April 2003 and completed the transfer and installation of 26 semiconductor tools for processing, measurement and quality analysis by mid-2003. March 2004 saw the full qualification of the first products from the fab.
The fab has seven metal organic vapour phase epitaxy (MOVPE) multiwafer reactors that provide the critical materials growth that is the foundation of all Bookham's InP devices. The final performance and reliability of the devices depend on the quality of the MOVPE material. The use of seven reactors allows the company to segregate processes - for example, heavily doped laser growths are not mixed with APDs on the same growth kits - and provides the flexibility and capacity to develop new processes.
Overall, Caswell has 5300m2 of manufacturing space, 3000m2 of wafer fab cleanroom, and a capacity of about 12,000 wafer starts/year. Its products include tunable distributed Bragg reflector (DBR) lasers, fixed-wavelength lasers (distributed feedback/DFB, buried-heterostructure and Fabry-Perot types), photodetectors (avalanche photodiodes/APD and positive-intrinsic-negative/PIN) and Mach-Zehnder modulators.
