This is a guest post by Philip Moriarty, Professor of Physics at the University of Nottingham
As Raphaël has pointed out in a previous blog post, scanning probe microscopy, although an extremely powerful technique which underpins a great deal of state-of-the-art nanoscience, can be notoriously artifact-prone. Not only is it important for the probe microscopist to ensure that the gains of the feedback loop(s) which control the motion and response of the tip are set correctly – and to understand fully just how the images are sampled and acquired – it is essential to realise that any SPM topograph involves a convolution of the tip and sample structure. It’s difficult enough ensuring artifact-free images on atomically flat surfaces at times, let alone on the curved surfaces of nanoparticles (which may have a radius of curvature comparable to that of the tip itself).
Before I get into a discussion of the specifics of the “striped nanoparticle” debate, let me add a few sentences about me. I’m a professor of physics in the Nanoscience Group at the University of Nottingham and my research interests span two key aspects of the debate: self-organisation of nanoparticles [1,2] and high resolution scanning probe microscopy (particularly dynamic force microscopy in ultrahigh vacuum and at low temperatures) [3,4]. Characterising and controlling the state of a scanning probe tip, and interpreting the associated SPM images, all play a major role in our research and occupy a significant amount of our time.
How does one go about ascertaining whether the features in an SPM image truly arise from structural, electronic, or chemical features at the sample surface, rather than an artifact of the scanning process? There are a number of very simple “tricks of the trade” that can be used to disentangle genuine surface features from, for example, vibrational or electrical noise; feedback loop ‘ringing’ due to improper gain settings; and/or tip contributions. Here are just a few:
- Like any other area of science, reproducibility is key. If the same area is scanned with the same imaging parameters, are the features reproducible? Do the features scale properly with scan area? If they are periodic, is the period the same if the scan speed is changed? If not, then it’s highly likely that a noise source (e.g. 50 Hz mains contamination) is ‘leaking’ into the image.
- If the gains of the feedback loop are too high (perhaps coupled with a scan speed that is too fast) the tip response will oscillate at edges. This effect can be exacerbated by unstable tips and can easily give rise to “ripples” or stripes in images.
- Depending on the relative radii of curvature of the tip and sample features, an inverse imaging process can occur whereby the surface images the tip rather than vice versa. Sometimes this is very helpful because it enables a characterisation of the tip (see  for a particularly elegant recent example) but this type of ‘inverse’ imaging has to be recognised and this is not always as straight-forward as it might seem.
So, do Francesco Stellacci and his colleagues’ images meet the criteria to be considered artifact-free? In my opinion, they don’t. This, of course, does not necessarily rule out the presence of stripes on the nanoparticles and, indeed, Francesco points to NMR and X-ray diffraction data which apparently support his case. Nonetheless, the evidence for stripes from the STM evidence is, at best, far from compelling. Raphaël Lévy and co-workers have raised a number of valid criticisms  which are not adequately addressed in Yu and Stellacci’s response . Instead of re-treading old ground, however, I am going to focus on difficulties with the data in Yu and Stellacci’s response as it is entirely symptomatic of the problems with their STM images in general.
I should note that Miao Yu and Francesco very quickly provided the raw data for Fig. 4 of their paper when I e-mailed them, for which I am very grateful. Note that this figure is a cornerstone of their counter-argument addressing Raphael’s criticisms. Let’s look at in detail because it highlights the key problems with their data acquisition and interpretation.
The figure in question is below. First, a large scale (80 x 80 nm2) STM image of the nanoparticle layer was taken (using, in this case, an ultrahigh vacuum STM) – this is shown in (a). Instead of then reducing the scan area to get a higher resolution/higher magnification image, Yu and Stellacci instead carry out what is sometimes called an “offline zoom”: they simply zoom in on a region of interest (shown by the blue box) using image processing software. This is a rather unorthodox strategy. The usual approach would be to zoom-in in “real time” on the area of interest.
Yu and Stellacci’s approach has some rather worrying side-effects. The image shown in (a) comprises 380 x 380 pixels and spans an area of 80 x 80 nm2. This means that their effective sampling resolution is 80 nm/380 = 0.21 nm/pixel. They then zoom into a ~ 15 x 15 nm2 region. This comprises 70 x 70 pixels and if these raw data were included in the paper it would look horribly pixelated. So the data is instead interpolated up to a higher pixel density. This acts as a low pass filter and spuriously introduces spatial correlations in the data. Seasoned scanning probe microscopists will note that the images shown in (b), (c), (e) and (f) below look to be low-pass filtered. This is not deliberate – Yu and Stellacci have stressed that their images are not filtered – but arises instead from the interpolation process applied to increase the pixel density.
An examination of the raw data (which Prof. Miao Yu sent me) indicates that the stripe features arise in many cases from as few as five pixels in the initial, un-interpolated data. It is clear that this is simply a fortuituous alignment of random noise. (Even Poisson-distributed (i.e. entirely uncorrelated) objects can often surprisingly appear correlated to the eye).
The question one must ask is why Yu and Stellacci adopt this bizarre approach to acquiring images of the ‘stripes’. If the stripes are indeed present, the obvious method to attain good quality images would be either to increase the pixel density in the large scale image (to say, 1000 x 1000 pixels) or, better, to maintain pixel density and zoom in. The offline zooming approach, with its associated interpolation and inadvertent filtering, is extremely susceptible to artefact generation.
These comments apply not only to the images shown in Fig. 4 but to the vast majority of the STM data produced by Stellacci et al. Throughout their work, strong claims of stripe formation are made on the basis of features which are no more than a few pixels across and which are extremely difficult to distinguish from noise contributions. Another prime example is Fig. 3 of Jackson et al., J. Am. Chem. Soc. 128 11135 (2006) where it is extremely difficult to ascertain even one period of the “stripes”, let alone make an accurate measurement of their periodicity.
Perhaps most damningly, however, is the lack of reproducibility. As I said above, a fundamental rule of thumb is that features should be reproducible from scan to scan (in the absence of significant mass transport due to diffusion – which Yu and Stellaci themselves rule out in their response) and scale appropriately with scan size. To the very best of my knowledge, Stellacci et al. have failed to demonstrate to date that repeated imaging of the same nanoparticle with the same scan parameters gives rise to the same stripe features.
Stellacci and co-authors will point to their detailed studies of the influence of scan parameters on the images of the ‘stripes’ they obtain. But their studies are fundamentally compromised by the lack of basic control experiments regarding the influence of noise and, moreover, their inability to show appropriate scaling of the stripe features as a function of scan size. I am nonetheless intrigued by the possibility of directly imaging the stripe patterns predicted by theory and would be keen to examine the particles using high resolution non-contact atomic force microscopy in UHV and at 5K.
Figure 4; adapted from Yu and Stellacci