Since several years so-called fast EBSD detectors are available. They mainly deliver a higher speed because of a clearly smaller native image size of the used camera so that the read-out time and the on-chip binning time is significantly shorter. This means, although the same interface for High- and Low-Res cameras is used, the different transfer rate for images of the same size (e.g. 160x120 pixels) is mainly controlled by the preprocessing on the chip. This trick increases the speed for a camera with VGA-resolution by a factor of 5 and more, in coparison to a High-Res camera. Although EBSD patterns are physically not that sharp, and practically patterns are always binned down to create orientation maps in a minimum of time, why EBSD detectors with a higher native resolution but remarkably slower speed are still the preferred detectors? Especially when speed is often declared as - but finally obviously not - an important driving force...
Higher resolution detectors can be made just a fast by binning, plus you have the option of using the highest resolution when needed (when the readout time is not a significant fraction of the acquisition time). These days high resolution detectors with say a 2k chip can acquire at 30 fps without binning. With the newer CMOS detectors the speed is not only much better, but the readout and dark noise is much lower than a conventional CCD which means you can acquire for a shorter time and maintain the same signal/noise. Win-win situation. The Hamamatsu ORCA-flash4.0 is a great example.
Hi Lawrence, regarding the CMOS you are certainly right. But why I should buy a HiRes detector only in case I perhaps need some high resolution images (how high, and what are you doing in case really see more than with 640x480 pixels)? Have you ever tried to save HiRes images for a map? Do you know, when your system starts to fail because of the transfer rate of the hard disc you have? :-) No, this is nowaday a very theoretical argument...especialy because ther is no commercial software available which is able to use this resolution (except of CrossCourt). All existing systems bin the patterns down internally (and binarise them!) in order to process them in an adequate speed, what is absolutely OK.
The ORCA-flash 4.0 is not used in EBSD, is it?
Greetings, Gert
PS: attached is a partial comparison between a Hires and a fast detector comparing mainly the [111] zone of Ni. Of course there are tiny differences but a) they are very exclusive for Ni, and b) what does it tell you? There is no commercial software available at the moment which takes advantage from this information...except of A. Winkelmanns dynamical pattern simulation program (similar code has been developed by Claire Maurice and Mark DeGraef).
I imagine it is only a matter of time before they are used for EBSD since they will provide greater speed and signal to noise. It is not a hardware problem for dealing with unbinned high resolution images but a software problem. I'm sure someone is working on software to use the added information efficiently (after all the spatial resolution of your detector determines the angular resolution of your orientations). With solid state hard drives you can transfer and store at a much faster rate than you would ever practically want to acquire images at. Already there are 4k CMOS chips available, and I'm sure it would be long before we get 16k.
At 30 fps the orca flash4 can transfer over half a million images (unbinned) via a USB 3 link to a solid state hard drive! It would take a little less than 4 1/2 hours before you reached capacity on a 4Tb drive. Obviously this is not the bottleneck. As usual it is the development of analysis software which is written to deal with the ever increasing density and of information which can be measured.
EBSD (50K) is no TEM (1+x Mill.) which is one reason that these detectors are not yet used for this applications. Simply too expensive and perhaps also too big (for the required geometry).
And...the information already now available in 160x120 pixel images is frankly speaking still not well used. So there is no real driving force, except for guys working on strain analysis, but there are still some other barriers like the inaccurate pattern center determination which benefits from a higher resolution first only theoretically. If you nowadays sit on an SEM and you are waiting 20min for the storage of a 6 GByte big file I don't want to argue what happens with the storage of 1 or 4TByte. Theoretically many things are possible but I can promise you that saving and processing big files is more challanging than you can imagine (all related to standard hardware and standard prices). And I want note that the fast detectors with lower resolution have been later launched...which had a simple reason. Even measurements containing all acquired patterns in a small format (typically saved as 80x60, or 80x80 pixel images) are nowadays the exception. Or do you save for EBSD measurements with patterns? In so far: Why people by hires detectors if they cannot use them (except for some simgle patterns which they cannot process but only publish? :-)
Thanks for your answer, Lawrence!
Gert,
For a 2k image each file is about 8 MB, and is readout and saved in milliseconds. I was just using an extreme example of the ability with standard hardware to continuously acquire, readout, and save 2k images at a rate of 30 images per second for 4 1/2 hours before you will fill a large hard drive. You can easily analyze thousands of images with software running on a single PC. It isn't the size of the files that would be a problem (reading an 8 MB file into memory from a solid state hard drive takes microseconds) but rather the type of computational algorithm that is being used. If you are just doing a least square fit to the data extracted (and the image analysis time is negligible) I doubt you would notice any difference in the time it took for the analyze to complete. If however you need to say invert a matrix extracted from the data then the amount of processing time grows exponentially with the size of the matrix (which is why for very large data sets no one uses direct matrix inversion methods and instead use back or forward projection methods and a large variety of other computation tricks for fitting extremely large data sets (and by data sets I mean the number of extracted values from the set of images not the image size). I use these cameras for x-ray diffraction. It takes on the order of seconds to process 100's of images (basically performing background and spatial distortion corrections to each image and then identifying spots and extracting their center of mass position, size, shape, integrated intensity). How long it takes to then fit the data is obviously independent of the size of the image files.
I am sure that in the near future EBSD machines will be using such detectors, and for the user there will be no noticeable difference in the time it takes to analyze each image, but there will be a noticeable difference in the precision of the analyzed results (higher angular resolution which as you noted would be extremely useful for those studying elastic strain). The cost of a CMOS detector can be as low as $8,000. For the higher end Orca-Flash4 they cost around $18,000, although that will continue to go down as even higher performance detectors come on the market.
Oh and I forgot about your question of size. The ORCA is about 85mm x 85mm x 125 mm and weighs 2 kg.
@ Lawrence: only one question: Are you doing EBSD in your lab? How much practical experience do you have with the storage of EBSD patterns (600 patterns or more per sec over 1h) inclusive the control of the SEM, and acquisition of EDS? This is at the moment the bottleneck...as far as I know (at least I should know it, cf. my profile). Regarding SSDs you are certainly right but you need to copy these data somewhere else and also this occupies the system you are using. I totally agree that there are always solutions already available somewhere, but the EBSD buiness is a) not that as big as XRD or EDS so that the required development ressources are quite limited, and b) the still existing technology is far away from beeing under pressure to become upgraded. We even do not use the delivered patterns properly. Why to increase then the image resolution? I understand your argument about orientation resolution but this is at the moment only of secondary order... since already the now available images are binned down even if you don't want this. This means, already with the images we get at the moment we can generate a clearly higher angular resolution. Don't ask me why this is not possible/available.
I wasn't pointing without any sense that the reduction in resolution is at the moment the trend for EBSD, not the increase. This is obviously a reaction of the market. Mor people are interested to measure faster, and they are satisfied with the currently available orientation accuracy. Moreover: More is not always better. You also don't buy now a bus for yourself, only in case you have the idea to travel in a few years with all your family members and relatives to a nice place. The costs for such an investment are simply not in relation to the profit.
I definitely don't want say that this will not change somewhere in future. My question was related to NOW. Nevertheless, also for future applications I am not very optimistic since the physical signal formed (proved by ideal (!) dynamical simulations) is not that sharp that it would require a higher pattern resolution. My pragmatic point of view is: We need to use the resolution which is necessary for the problem and not which is technically possible (especially if it is much cheaper).
Regarding size and weight: the cameras now used have an estimated weight of about 50-200g and the size is in the scale of a matchbox. Unfortunately the lens and the high-precision moving systems are quite costy and heavy so that these detectors look clearly bigger. The best way would be anymway a direct imaging without screen, but also there some very specific problems need to be solved exclusively for EBSD. Up to now, EBSD mainly took advantage from general developments in IT and digital cameras technology. These were no genuine developments of EBSD.
If you need to take 600 fps (I was not aware that you could get sufficient signal to noise with an exposure of less than 2 microseconds. I am also amazed that you can readout the image quickly enough to allow such speed. Is this an inline readout chip which allows zero dead time?) then I would say that the current detector technology is about 3-5 years away from allowing the increase in pixels and decrease in readout noise that I was talking about at those speeds.
The maximum speed of binned (!) images was 6 years ago already in the scale of 1000 patterns per sec (CCD chip). With a hit rate for Ni about 98%. The required current is then about 15nA (at 20kV). With 30kV you perhaps need only 10nA.
Ok, well I think that explains it. There is always a trade off between speed and resolution. Clearly speed trumps resolution for EBSD applications. Even at a synchrotron the signal to noise requires acquisitions times in the range of 100 ms to seconds so resolution (which is basically the same thing as the number of pixels since the field of view can be modified with optics) is the limiting factor. With the development of the 4th generation free electron laser sources, though, the signal can be orders of magnitude higher which means that speed becomes an issue for measuring extremely fast kinetic processes. This may be what is pushing CMOS detectors towards faster and faster frame rates. I think the current technology is at about 100 fps for unbinned 2k images, but of course this changes rapidly. Just a few years ago being able to acquire 2k images at 10 fps was pushing it (for reasonably priced detectors). It should be interesting to see where the technology is a few years from now.
Already a few years ago I tried to change to CMOS for EBSD but it is not easy to go this way if there is no real need for it. I assume that if CMOS will in general replace CCD also EBSD will make this step, except some totally new application requires it. The EBSD market is (still) too small and the few people too busy with more urgent tasks that forward-looking projects are often of smaller priority. Simply too risky if such investment fails... Structure analysis by EBSD could change this :-)
Excellent information Gert & Lawrence!
2k image means, 1000x1000 or 2000x2000 pixels? In XRD, the most common challenge (bottleneck) is SNR (signal to noise ratio) in laboratory based XRD systems.
"Even at a synchrotron the signal to noise requires acquisitions times in the range of 100 ms to seconds". Wow! For what sort of experiments?
I can only speak for the EBSD business :-). In order to keep the prices low for the customers the detectors are based on common cameras with use CCD chips from mass production, e.g. 800x600 pixels or 640x480. Some detectors are also based on unusual formats. At the moment either 640x480 chips are used for fast detectors (all EBSD manufacturers), or 1024x1024 (EDAX-TSL), 1376x1024 (Oxford Instr.), and 1600x1200 (Bruker Nano) for the so-called High-Res detectors. As already pointed out, commonly a binning is used to improve the quality (reduction of noise). Comparing the images no real reduction of quality becomes visible (except for the strain application for "HiRes-EBSD"). In papers often the binning is given as only information but this does not help that much since you need to know the native resolution of the detector. It is typical "lab slang" since internally people knwo which detector they have.
In Lawrence case I assume that 1k means: 1024x1024 and 2k respectively 2048x2048 pixels.
What is the typical cost of the entire EBSD upgrade? What is the typical cost of the base equipment (hardware) with necessary software? Today?
@Ravi: This hardly depends which configuration you are selecting, how many licences you need and which discount you are able to negotiate. The detector prices are usually identical or only slightly different. In fact only the cameras and the lense systems are different. I guess realistic prices are about 60k€ and upwards. Special offers are possible but then be sure that you really get all whyt you want. The EBSD manufactureres also have to survive! You might be happy at the first time but perhaps you will realize later that the hardware is perhaps one last system before a new has been invented, or you are missing software options which you believed to get but they have been presented but are now not included. A complete system with everything has perhaps a price about 90k€ incl. discount. These are really only estimations and might be not very realistic. It is better to contact a local dealer :-).
Coming back to my initial question I am still wondering why people still order a high resolution detector (although they cannot use them...if we exclude for a moment the HiRes-EBSD guys which moreover need an additional software package, e.g. CrossCourt) if speed is nowadays the only driving force, and regarding resolution both detectors deliver the same information (because of limited software features)? Is it only a marketing gag, or do people prefer "to buy a bus instead of a compact car" because they don't have to pay for it personally?
Good stuff José M. Amigó! Thanks!
Excellent Gert! That gives me a lot better idea. Thanks!
- Badges
- Science topic
- Similar topics
- Physical Sciences
- Materials Science