Isn’t it time to get serious about standards?

Devices made from compound semiconductors have at least one key difference to those of most silicon-based devices – they vary in chemical composition from layer to layer. So, given that, you might believe that accurate measurements of composition would be a valuable endeavour and standardizing those measurements a no-brainer. Take that route and, if you work on behalf of an RF chip manufacturer with three different epitaxy suppliers, you would then get the same Al0.18Ga0.72As layer in every single chip you buy, regardless of its vendor. Or if you send your defective chips out for SIMS (secondary ion mass spectrometry) analysis, you would get data that indicates the same alloy composition from your outside service as you do from your in-house diagnostic instruments.

That all makes a great deal of sense, but it is a far cry from the world we currently live in.  Instead, what really happens today is that most epitaxy companies and analysis companies draw on individual reference materials.

There is a huge stash of them on the shelves of these firms, because it’s often not just one reference per customer, but one reference for each device process.

What this means is that efforts at determining the composition are closer to resembling the matching of paint colour cards than the making a scientific measurement. In practice, engineers tend to add a little bit of this and a little bit of that until they get a layer that matches the one they grew last month.

Several reasons can account for the persistence of this inefficient situation. Habit probably tops this list.  Take a compound semiconductor alloy system with a long history of technological importance: AlxGa1 - xAs. 

This alloy has the extremely useful property that its lattice parameter has almost no dependence on the aluminium mole fraction. Thanks to this very favourable attribute, hetero-junctions can easily be grown without introducing strain relaxation from defects. This has led to widespread determination of the aluminium mole fraction with an X-ray diffraction (XRD) rocking curve that can uncover the small elastic strain in AlGaAs.  But this method is not flawless: It can be fooled by changes in the substrate lattice parameter or by a doping-induced expansion or contraction of the lattice.

Fortunately, this is not the only approach to determining AlGaAs composition. If the aluminium content is low, its mole fraction can be uncovered by measuring the band gap energy with photoluminescence.  This is a high precision method, but it is again subject to systematic errors from sample heating, impurity and doping shifts of the apparent band edge, and excitation intensity shifts in band-edge transitions.

If you have learnt to determine AlGaAs composition of an epilayer, what method do you follow?   Chances are that it is one of the two listed above – measuring strain relative to the substrate with XRD, or determining the bandgap with photoluminescence.  And it’s probable that you convert this number to a composition with an equation given to you by your thesis advisor, an older graduate student, or your supervisor.  If you changed institutions, you might have taken your equation with you, or maybe you accepted a new one.  The local nature of these conversion factors is often an impediment to the absolute accuracy of any composition measurement.

Up until now, we’ve only discussed the problems associated with ternary alloys. These are magnified to an entirely new level with quaternary compounds like InGaAsP. When the National Institute of Standards and Technology (NIST) sponsored a round-robin comparison of this class of material in 2002, this institute found substantial variations in XRD and photoluminescence data on identical samples examined in nine different laboratories.  Don’t put down these variations to laboratory-dependent calibration factors – they persist even when these factors are removed from the data analysis.

X-ray diffraction is widely used to determine the composition of aluminium in AlGaAs. Care is needed, however, because interpretation of the data must account for the substrate lattice parameter and doping-induced expansion or contraction of the lattice. Photo by James Burrus, NIST

Fortunately, it is not all doom and gloom: We have the technology to remedy this situation.  Starting in 1997, NIST began a programme to standardise the measurements of compound semiconductor composition, starting with AlGaAs.  Some of the impetus for the work came from the Optoelectronic Industry Development Association (OIDA), and NIST sought input through venues such as CS-MAX and SEMI committees. 

Methods for determining composition analysis were examined during this programme, with several papers published that quantified the measurement uncertainty for those methods and outlined best practices. In 2006, the programme culminated with the production of Standard Reference Materials (SRMs) for an aluminium mole fraction in AlGaAs near 0.20 (SRM 2841) and near 0.30 (SRM 2842). Each standard consists of a layer of AlGaAs about 3 µm-thick on a GaAs substrate. The AlGaAs layer has been certified to have a stated aluminium mole fraction with a typical absolute uncertainty of 0.002 (2 s).  

Another highlight of this programme is that it led to a refinement for the correction factors for aluminium, gallium, and arsenic in the CITZAF method for the accurate interpretation of data collected in electron microprobe X-ray analysis. 

More recently, the range of materials has increased, with NIST producing SiGe composition reference materials (RMs) in response to industry requests.  (Without getting into too much detail, the RMs differ from the SRMs in that they are stated to be suitable for their intended purpose but not directly traceable to the mole.)

One of the benefits of having reference materials available is that they can be used to improve accuracy of other compositional analysis methods, such as Auger spectroscopy, X-ray photoemission spectroscopy, and SIMS, provided that any necessary corrections for sampling depth considerations are included.  The Fundamental Parameters projects – efforts by X-ray diffraction companies to calculate composition correction factors from fundamental parameters – is in fact one example of an attempt to provide sampling depth and matrix corrections to composition measurements dependent on X-ray emission and absorption. If reference materials are not employed, these methods tend to produce data with clear systematic errors, such as deviations far from 1.0 for the (Al+Ga):As ratio.  But if a good set of standards is used, accuracy improves to where it matches or exceeds what can be achieved with XRD and photoluminescence.

So, how do we move forward?  It’s simple: We just have to use what we have!  NIST has been reluctant to pursue further development in new alloy systems, such as the group III nitrides, because the SRMs have remained mostly on the institute’s storage shelves. 

However, there has been a successful implementation of their use at Evans Analytical Group, where they have been employed to calibrate SIMS measurements of a AlGaAs-based HEMT structure to aid manufacturers in identifying shifts in epitaxial supplier composition.  All that it took was a little investment in an SRM artefact and some comparative studies to extend traceability. 

Photoluminescence is a widely used method for determining the aluminium composition in AlGaAs. Photo by James Burrus, NIST

At a recent gathering, the 2013 Lawrence Workshop on Standards, this kind of work was identified as being of high value. The workshop also highlighted communication difficulties that can create a barrier between standards suppliers and standards customers.  To help to address this, NIST has on-line resources to guide interested parties in how traceability can be documented.  If industry starts to use these resources and reference samples, we can move on to a time where composition determination is a precise science, rather than the crystal-growth equivalent of mixing paint.  

·         Contribution of an agency of the US government; not subject to copyright.  Certain companies are mentioned by name in this article. This does not imply endorsement by NIST; other companies may provide similar services of equal or greater quality.