There are several SEM operational parameters (such as working distance, probe current, beam diameter, accelerating voltage) that can be considered primarily. Once these parameters are standardized then you can look for counts.

Also u need to consider characteristics of ur sample (homogenized composition or not, in case of second phase, its size and type of elements u are analyzing etc.)

This is just a crude way of looking at it......

First of all you have to consider if you wanna perform a standarless analysis (or semi-quantitative) or a quantitative analysis. If you really need to know the composition of your sample with an error of the 1-2% element by element, go on with quantitative analysis with standards. If you are just interested to recognize different phases a semi-quantitative analysis is enough. Furthermore, you have also to consider, depending on your samples, if it´s the case to use WDS.

A very important parameter will also rely on you: your sample must be very well polished and the exposed surface must be as flat as possible. Otherwise there is no "good" or "bad" SEM to my knowledge. However, as outlined by Ruggero Vigliaturo, things will become more tricky if you begin to work with standards.

There are several differences in EDS detectors that could be important for your analyses. For example, state-of-the-art EDS detectors can have different energy resolutions, in the case of SDD detectors somewhere between 129 - 121 eV. This becomes important when you want to analyze compositions of many light elements. With a lower (better) energy resolution you are able to prevent many overlapping peaks. Also, for speed of EDS mappings the sensor size is important. There are EDS detectors with sensor sizes between 10mm² and 150mm² or even larger. So if you plan to acquire high-quality maps you should look out for a large size detector.

1. The most important – to work with the best operator. EDS is capable of many things, unfortunately most EDS operators are not capable to use EDS properly.

2. If you need quantitative analysis, you should ask whether the lab has certified standards (standarless analysis in most non-metal specimens is useless).

3. If you need light elements, ask if they have ultrathin window.

4. Newer machines are better (but only if they are served by good operators.)

1. Do you need mapping or not? If you need, you’ll have to ask for a SDD detector.

2. Do you have samples that are charging in usual SEM conditions? If you have, than you’ll need an environmental SEM or at least one with low vacuum capability.

3. What precision you need for your analysis? If 1 % errors are enough, you can do it with standardless analysis. If the errors should go to 0.1%, than consider standard quantitative analysis (you should ask for proper standards availability). If precision have to go under 0.1%, try to consider other analysis (WDS, XPS, etc.).

4. Do you need light elements analysis? If so, ask about the detector limits (Be is the lightest usually for ultrathin window detectors).

5. Keep in mind that precision depends on sample’s flatness, composition (some elements are not discriminated by EDS), conductivity, element concentrations, etc.

@Catalin

1. Any detector is good for mapping (SDD is better, but not at all required)

2. Environmental SEM is not really good for EDS, so to prevent charging better to go with carbon coating (or even thin Au coating)

3. You are too optimistic regarding standardless analysis. Its precision is much worse and for non-metal specimens it is mostly useless. Even with standards you cannot expect precision better than 2%.

@Vladimir

You are right in most aspects, but a bit too pessimistic...

1. When you have a Wolfram filament with an average lifetime at 50 hours, and your Si(Li) detector will take over 10 hours for a 600x480 pixels map, than you’ll understand my point.

2. I am using low-vacuum successfully (up to 60 Pa) with a conical option for lowering the skirting. At larger pressures the differences start to be significant.

3. EDX has many limitations, but I gave the best limits. It’s usual to have larger errors, sometimes 10x larger, e.g. for glasses.

If you really need to know the composition of your sample with an error of the 1-2% element by element, go on with quantitative analysis with standards. Operator's skill is vitally important to get dependable result. You may try your luck with the SEM-EDX Lab, 2nd Floor, Central palaeontology Division, Geological Survey of India, 15 Kyd Street, Kolkata-700016, West Bengal, India. Its a Govt. institution and Dr. S. Shome is trained abroad twice for operating the latest Zeiss SEM and EDX. They have good set of standards and are carrying out excellent jobs. They do out side jobs on payments. You contact Dr. S. Shome, Director, +919836250888 for further details. Good luck.

Thickness of the film is also a parameter for a good EDS result. Characteristic X-Rays are emitted above 1 micrometer thickness. Therefore your film must be at least near this value.

With a direct answer to your question, one would need more details to provide an accurate and useful answer tailored to your cause. There is no definite and optimal parameter. Every function is important. Are you doing qualitative, quantiative or both? If going for quant, what type of quantification do you need? What type of sample do you have? Is it homogenous? And what is the purpose of your analysis? The answers to these questions should help in identifying optimal parameter usage etc.

@Matthew 

I am asking for wide range of samples which are used for different projects so I was looking for general answers. 

As ususal there is no general answer although the opinion of Vladimir Dusevich is probably the most relevant: you need to look for a very experienced operator. 

Nevertheless, a few comments even if there are some repetitons to former statements.

a) the SEM is usually not that important since the final X-ray probe is more in the dimension of several 100nm or microns which is usually fullfilled even by W-SEMs. It is clear that for higher magnifications FEG-SEMS (Shottky filament) are better because of the higher electron density within the beam. Nevertheless, also with W filament you can achieve a quite high beam current (with lower spatial reolution).

b) The state of the art are nowadays SDDs (silicon drift detectors). They have certain advantages but also disadvantages. The bigger size (50...150mm² in comparision to the classical 10 mm² Si(Li) detector chips) allows to collect much more counts (please consider that the covered solif angle is important and not the detector size!), however, the electronic behind is hardly working on the limit. This has the effect that you can reach the given energy resolution (which is given for Mn radiation!) at low count rates since the electronic cannot seperate the incomming electrons with the same accuracy. For mappings where the elements are given by peaks with clearly different positions in the spectra this is not problematic. Here you can somtimes operate with 1Mio counts/sec. But please take into account that not the displayed incoming counts are important but the usable output (considering something like a "dead time" which is no dead time in the classical meaning but simply expresses how much electrons could not be reliably assigned a certain energy). This means, here size doesn't matter if the incomming signal cannot be processed (and there are real differences between the manufacturers (cf. c) )

If you go to light elements (low energy range) where you naturally have numerous overlappings you need to spend more time because of the reduction of count rates (caused by lower accelaration energy). There this data overflow does not play any role so that the given "dead time" is practically zero. Here size matters!

c) As far as I know there are only two manufacturers of SDDs worldwide, i.e. practically all EDS manufacturers use only these two chips. Only their data processing is different (automatic background processing, escape peak processing etc.). The two different SDD types are different in their designs which have certain advantages and disadvantages as well. Thus, only one type can be cooled down quite easily by a multiple Peltier cooler, and the cooler the SDD the better the reachable resolution.  

d) The nowadays preferred standardless quantification has been developed since the alternative would be a collection of element spectra for different aquisition conditions. 10 years ago EDS could be only operate at 20kV since only for this acceleration voltage a "complete" set of spectra was available. If you want measure at 10 or 5kV a customer has to collect standard spectra by himself, what was and is not very comfortable. in so far I would compare the situation for EDS with that of XRD where 20 years ago for the quantification standard mixtures were required. Nobody would do this nowadays anymore although it would certainly hepl to increase the reliability of quantifications of specific mixtures of phases. However, practically there are much more relevant sources of errors which are not covered by these standards.

e) The reliability of a quantification depends on many factors. The most critical is the composition itself. If there appears a strong overlapping of peaks the reliability is much worse than for a phase where no overlapping appears. Moreover, the visible element-specific signal is different, so that some elements can be described with a high accuracy in comparison to some other, like carbon. This is also relevant for the radiation used. A K line will give a different (composition)  result than a L or M line since their detection accuracy but also overlapping with lines of additional elememts is different. This is only one reason why Vladimir underlined the role of the operator. The software always produce some numbers but if you change the sample composition the reliability of the quantification changes (but is NOT pointed out by the software), and it needs a lot of experience to consiser all affecting parameters like sample preparation. Of course you can measure at rough surface but if you believe the quantification results it is your fault! What is rough? This is a totally underestimated source of error since the signal does not come automatically only from the arae you are scanning with the primary beam but from everywhere where backscattered electrons hit anything (the sampel holder, another sample, the stage, the chamber etc.) and it is part of the solid angle covered by the detector collimater.

f) other sources of errors are e.g. local sample charging, low chamber pressure, coatings, detector window (pollution like ice or oil). 

Finally I would suggest to use some standard materials which are close to the material you want analyze and send them to a lab. From the results you can judge whether they work properly for your demands.

Thanks for everyone, this question and the subsequent answers helped me a lot. I would kindly ask something else in the sequence of this discussion: what would be a good parameter of quality in an usual EDS mapping image? Often, in published articles, is difficult to find some parameter setup to ensure a good quality EDS analysis. What would be the counts/pixel best value? Is it a good parameter, or it will depend on the type of machine and sample you have? Is there some kind of mounting stage, geometry or positioning that can increase the counting rate?

hello everyone, there are two different sample one of them includes Zn and the other one includes Co as same weight, when i do eds the one which includes Zn is okay but in the one which includes Co there is no Co can be found by the EDAX . I have researched that it can be about differences between their masses because in the sample there is 1000:1 PAN and ZnPc or CoPc . Is there anyone who can explain why Co cant be seen on EDS analyze, thank you