MBE Dreams: Dream Software

Faebian Bastiman

Returning to the theme of “Dream MBE”, I would like to spend a moment discussing molecular beam epitaxy software. Off the shelf MBE software is passable at best. Ok, so it can open and close shutters, perform temperature ramp ramps and execute a sequence of commands in some form of recipe: This is, when all is said and done, the bare minimum. When you spend some time actually considering the features you would like in MBE software, you quickly realise nothing on the market really hits the spot. So… I have spent the past six months writing my own MBE control software: EPIC™.

You cannot create the world’s greatest MBE software in six months, however you can implement an exciting list of features and lay the foundation for fantastic things to follow. Beyond the bare minimum functionality, the features of EPIC™ are:

  1. Comprehensive, optimised software-hardware interface (SHI)
  2. <1 millisecond to process a shutter request
  3. Full device parameter explorer and editor
  4. Ability to monitor PID loops, pyrometers, plasma sources, pressure gauges, pumps, valves, motors, services (LN2, water flow, gas pressures) status and other hardware (e.g. reflectivity data) all from a central, grouped, tabulated status screen
  5. Associate any number of physical devices with an individual control loop
  6. Ramp to temperature set point or output power for PID controllers
  7. Flexible, simple script recipe with full control flow, local and global variables, unrestricted parallel editing and execution including “edit on the fly” and access to any parameter from any device
  8. A colour co-ordinated message box
  9. Wake-up and sleep scheduling
  10. Substrate location and status tracking
  11. Automatic flux scheduling, gathering and Arrhenius/polynomial fitting
  12. Atomic flux, growth rate, doping and composition calculation tools
  13. Material usage tracking and cell fill level estimation
  14. Two data logging modes: SPAM (log at fixed interval) and SMART (log only on change)
  15. Watchdogs warnings and actions
  16. Integrated simulation mode with full hardware emulation

Ok let’s analyse that numbered list. (1) SHI is a fantastic interface with every device possessing its own read and write thread. SHI enables (2) and (3) by its very nature. Not every process requires less than 1 ms shutter response, though since it exists by default I thought I might as well include it in my list. The device explorer is also a nice feature, since the front panel menu on most devices can be a pain.

The status screen (4) is very comprehensive and the layout facilitates ease of reading. A quick glance can reassure you that every part of the MBE system is ok (or equally quickly warn you when it isn’t!). All necessary items are interactive, giving the ability to perform ramps and open/close single/multiple shutters. The ability to associate any number of devices with a control loop (5) is very useful for plasma sources when they may have: a mass flow controller, a RF power controller, a main shutter, several valves and an optical monitor. So too is performing power ramps (6) for certain control loops, like carbon sources.

The recipe script (7) is essentially its own programming language complete with interpreter. Each recipe executes on its own thread and all commands run to 1 ms precision. The control flow allows you to step through each line individually one-by-one or allows the recipe to run freely, you can pause any line, go to any line from any line, skip lines, stop the recipe at any point and edit the recipe on the fly (only when paused for safety) and the recipe checks every line for errors on edit. Local and global variables allow parallel recipes to control complicated processes. Whereas access to any device parameter gives the ability to do almost anything, including in situ reflectivity control (see Dream Machine).

The message box (9) simply keeps you informed with a variety of colour co-ordinated, time stamped status updates including “Recipe X completed”, “Ramp X started”, “Sample X moved to position Y” or the less pleasant but equally useful “device X failed to respond”.

Wake up and sleep is fantastic (9). One can sleep much better knowing the machine will wake up at 06:30, do all the flux checks for the day and wait for you to stroll in at 09:00. Equally, the sleep mode prevents that troublesome “did I ramp down the Gallium before I left?” since it sets the system to a predefined standby state.

When you have more than a handful of substrates and/or more than a single user it is very useful to have a substrate tracking screen (10). Each substrate can be individually tagged with a name, a colour co-ordinated status (new / degassed / grown) and its current location in the system. The substrate tracking paves the way for extended functionality such as automatic run logging, automatic growth sheet creation, automatic sample transfer and batch sample processing (all of which are on my to do list).

Any software that purports to convert effusion cell temperature into growth rate is just plain lying! Beam equivalent pressure (BEP) is typically what you measure and atomic flux in atoms/nm2/s is what you get. Growth rate is simply something you induce depending on the atomic density of your substrate. Software that gathers flux data for you (11) and plots it on a graph is a welcome start. A one click fitting option ranging from simple Arrhenius to 12th order polynomial fitting is sufficient to handle all source and opens the door for much, much more. The first step is enabling conversion from BEP to atomic flux. Once done you can use a growth rate and doping calculator tool that updates to the correct doping density and growth rate for any substrate (12). You can then “ramp to doping density” or “ramp to growth rate” rather than “ramp to set point”. You can estimate the real time composition during growth and calculate the composition ahead of time for a given cell temperature combination. Perhaps my personal favourite is that you can integrate the flux with time every single second and track the cell material usage, display it on a simply bar chart and estimate the fill level of each of your sources to 1%. All of these flux related features are present in the current version.

Of course you want to log all relevant data values (14) and the option to log only on change prevents large data files being created when the system is idle (over a weekend). You also want to monitor the instantaneous data values and be warned when they are outside of an “OK” range. Moreover, for certain values you would like the software to take an action on your behalf to prevent damage (15). Finally you may want to run the software in “off line” mode to “try before you buy” or even to “test a recipe before you run it for real” and in these circumstances the simulation mode is very handy.

Q: What could possibly be missing?

A: Access to the source code.

The reasons for open source MBE software are almost as numerous as the feature list itself, so let’s stick to the two most important ones:

  1. Add device drivers for any hardware
  2. Add/edit/remove content to the user interface

Adding device drivers is somewhat essential. The system hardware changes over time as the system is upgraded or re-commissioned for a new purpose. The current EPIC™ has a generic driver template that can be used to create a new device driver in around 30 minutes. The next step is to make that driver template available to all and to use device driver DLLs. Fairly simple.

The user interface is a very personal thing. The software-hardware interface (SHI) is rather universal once done right, but the user interface is a matter of taste. Script recipes are very powerful but perhaps not for everyone. Tabulated data makes for fast referencing but perhaps diagrams are more your thing. Etc. Etc. Etc. There are three solutions beyond making the software open source:

  1. Create a software development kit (SDK) for EPIC™ to allow everyone to customise it with their own desired features
  2. Allow SHI to be a stand-alone data acquisition tool that can be used either via windows API or via a socket interface to allow the user to create their own graphic user interface or web application
  3. Provide customisation on demand and provide a tailor made solution for each user

Happily EPIC™ will have all three.

One final question: Where can I try the latest EPIC™ beta demo version?

Here.

Advertisements

MBE Dreams: Efficiency of Effusion Cells

Faebian Bastiman

Have you ever wondered what the material delivery efficiency of standard MBE effusion cells actually is? We typically spend on average £2k a year on source materials and £1.5k a year on substrates. We slice our substrates into 11.4 x 11.8 mm2 pieces and get 12 from a 2” wafer. This means our substrate usage efficiency is around 80%. This is the price we pay for cleaving square pieces out of circular substrates. This means we are using £1.2k out of our £1.5k of wafers and losing some £300 a year directly into the recycling bin.

But what about our £2k of cell material?

Well we buy a 240g charge of As every 3 years. This means we load 1.93 x 1024 atoms into our system every 3 years. We grow on average 1 µm a day, on a ~10 x 10 mm2 active area, 280 days a year for 3 years. This is 84 mm3 of GaAs or 1.86 x 1021 As atoms. That means out of 1.93 x 1024 As atoms we put into our system only 1.86 x 1021 actually end where we want them i.e. in our epilayers. That is 0.1%. Ouch, 99.9% material wasted!

Ok, just breathe. We know As is pretty wasteful, it is so gaseous that it goes everywhere. What about our group III’s?

Well we buy a 30g charge of Ga every 6 months. This means we place 2.6 x 1023 Ga atoms in our small 10cc effusion cells every 6 months. We still grow the same 1 µm a day, 87% of which is Ga. We grow this for 140 days in 6 months on our same ~10 x 10 mm2 active area. This is 12 mm3 of GaAs or 2.7 x 1020 Ga atoms. That is also an efficiency of 0.1%.

So out of our £2k a year on material we are only using £2 usefully. The other £1.998k ends up plastered all over the chamber walls, caking shutter blades and gathering in a large pool at the lowest point on the system.

Can we do anything to improve the situation? Sadly for standard effusion cells the answers is: very little.

We could reduce the cell-substrate distance. In the extreme case you could have the sample at the cell orifice, however the temperature rise when opening the shutter would be horrendous and group III cells tend to spit material too. The cell-sample distance is related to the number of sources, and assuming you want 10 cells and a pyrometer pointed towards your sample the current distances are already the lowest possible.

We could reduce the cell’s orifice with a Ta aperture. Effusion cells are designed to create a uniform flux across a certain sample area however they also spray material non-uniformly in almost every direction. In our case the sample area is 4x less than intended, so we could potentially achieve 4x the efficiency (a jaw dropping 0.4%) by reducing the orifice. The minimum aperture is related to the required flux uniformity, and in most cases this is optimized at the time of manufacture too.

We could also lower the deposition rate. There are some fixed time intervals for MBE, like the time the shutters are closed whilst the oxide is being removed and whilst samples are being transferred. When you grow at 0.1 ML/s you are losing 10x less material than when you grow at 1 ML/s in these fixed times. Of course some times lowering the deposition rate does not help. When you are growing a 5 nm / 5 nm GaAs/AlAs superlattice, for example, it does matter what the deposition rate is since (assuming  the growth rates of the two cells are identical) the atomic equivalent of 5nm of Ga will be deposited on the Ga shutter whilst it is closed during the 5nm AlAs layer, and vice versa. The efficiency increase depends on the ratio of growth time to preparation time. If we grow a 100 period 5 nm / 5 nm superlattice at 0.11 ML/s we waste 0.37x less material than if we grow at 0.55 ML/s but the total sample growth is increased by 1.9x. This comes down to throughput, the data in this article is gathered for a deposition rate of 0.55 ML/s, 0.000001 ML/s is very efficient but… 

With the orifice and the 0.11 ML/s growth rate the efficiency of a Ga cell would be 1%. This is certainly better, but still extremely wasteful.  

The final option is to used valved sources. Whilst valved sources are not particularly effective for high vapour pressure elements like As and P, Sb can be delivered with much higher efficiency because it behaves more like an Architypal molecular beam than a gas source. Valved sources are pretty standard for group Vs but what about group IIIs? Well e-Science are currently introducing their Valved Titan effusion cells for Ga and In. A valved source has two distinct advantages over a standard cell. On the one hand the material wastage is significantly reduced, since the cell can be idled hot with the valve closed. On the other hand, varying the valve position can enable instantaneous changes in deposition rate. A fully valve sourced III-V MBE system is certainly amongst my MBE dreams.

Molecular Beam Epitaxy: Dream Machine

Faebian Bastiman

I want to start this article with a question. What do you not like about your MBE system? Take a moment and think. I am certain at least one problem will immediately materialise within your mind. Indeed, after teaching MBE operation to a number of PhD students, I constantly meet the question “Why?” “Why is it like that?”, “Why is that so slow?”, “Why is that not automatic?”, “Why is that not controlled?” and possibly my favourite: “Why do I have to come in on a Sunday evening to do that?” The reason MBE is “like that” is that we, the users, are “putting up with it”.

There has been tremendous advancement in MOCVD/MOVPE in recent decades because there has been a demand from industry. I think it is high time we, the researchers, made similar demands of MBE. Industry and academia are very different environments, but both adhere to a basic principle that I term “money-time” duality.

Consider a typical R&D MBE system. Sample transfer is an art form. Growth rate calibration is tedious and laborious. Instability hinders the systematic. Maintenance is complicated, laborious and hinders productivity. Throughput is manpower not machine limited. One skilled operator can probably grow two good samples a day. Research progress is, inevitably, slow. Bear in mind one system is probably facilitating several projects, collaborations and multiple characterisation studies.

Let’s analyse the above paragraph. I have actually operated a few MBE systems where I drew a great sense of achievement from actually successfully transferring a sample to its destination. Should sample transfer be the most challenging and rewarding activity in MBE operation? No! No it should not. So let’s automate that straight away. Automatic sample transfer is a must. It also opens up the possibility for batch processing: the execution of 12 samples without user interaction. Detailed systematic studies take days rather than months. Productivity has increased 1000%. Excellent, what next?

Growth rate calibrations are time consuming. I have operated a system where it was basically 90% of my job to monitor the cell fluxes, recipe writing the other 10%. So let’s automate that. Automatic cell flux tuning before every recipe negates the need to have user intervention and reduces instability. The growth rate can be calibrated in several ways. Until recently I considered RHEED the most valuable, now I believe in situ reflectivity to be even more powerful. With reflectivity feedback, sample temperate, layer thickness and composition can be controlled to <1% deviation. Suddenly MBE is highly systematic and stable. So let’s have automatic in situ reflectivity controlling the sample recipes. How about a dream software suite to go with out dream MBE system? Yes please, next?

Maintenance will always be a problem in MBE operation, but it need not be a crippling blow. Part of the maintenance responsibility falls to the user establishing good MBE practice but equally part of the responsibility falls to the manufacturer. There are several items that fall into the category “ease of maintenance”. Ten minutes (or perhaps an hour) talking to an experienced technician would highlight hundreds of issues that could be designed out. Maintenance should simply be regular, straightforward and swift. Let’s pay attention to the people who have spent their entire lives maintaining and fixing MBE system and incorporate their ideas, ok, what else?

One of the hidden problems is unit cost and this certainly limits the number of MBE research groups. Skill is not a limiting factor. A good teacher can teach the basics of MBE operation in about two weeks. A good student can become autonomous in about 3 months. Of course here we have a dangerous situation: if the system is fully manual, the process will take longer but the student will gain a deep knowledge of MBE. On the contrary, if the system is fully automatic the process will be much swifter, however the MBE knowledge gained will be more superficial. It is therefore always best to learn on a fully manual system. The automatic system is however essential for any serious research. This “dream machine” is not a wild fantasy. Everything mentioned in this article already exists. It is simply a matter of putting the pieces together.

So the final and most important question: How much will this system cost? Well before I answer, I will ask you: what is the cost of not having it? The answer is fewer research groups, less samples, less science, less progress, less understanding, more downtime, more maintenance, more frustration, more money wasted maintaining out of date machinery. The problem with the dream machine is not the £100k of parts; it is the £400k we pay to have those parts assembled. Much money can be saved by stripping the useful bits off old systems: the pumps, valves, cells, flanges, nuts and bolts and the racks. What we need is to invest a little time and effort into creating a compact, inexpensive stainless steel chamber and supporting frame integrated into automation software. A nice RHEED, reflectivity system and an As cracker. The substrate heater need only accommodate 2” or ¼” of 3” wafers maximum, we are performing R&D after all. 10 cell ports are plenty, by the time you have grown 360 samples in 1 month you will have thoroughly exhausted the cells, thoroughly explored one material system and can load a different combination of cells for the next month’s work.

How much does the system cost?

£500,000 new

Growing 4320 world class samples a year with less than 15 minutes work a day?

Priceless