The Importance of Advanced Scanning Technologies in Microscopy

Scanning microscopy has been around since the early 80s, but many groups, including those at Nottingham, are still struggling to achieve optimal results. One major issue is the quality of the probe used, which can lead to frustration among researchers. A good tip is crucial for high-quality imaging, and it's not uncommon for researchers to spend hours trying to optimize their probes before achieving satisfactory results.

Training a Convolutional Neural Network

Recently, a breakthrough in scanning microscopy has been achieved by training a convolutional neural network (CNN) to distinguish between single atom tips and double tips. This is a significant improvement over previous methods, which relied on manual inspection of the probe's structure. The CNN was trained using a large dataset of images of probes with different structures, allowing it to learn the characteristics that distinguish a single atom tip from a double tip.

The Next Step: Optimizing Probe Design

But what's truly exciting is that this technology has been taken further by modifying the probe itself to recover its optimal structure. This means that researchers can now control the tip's shape and material properties to achieve specific imaging conditions. For example, if a researcher wants to image a particular chemical bond or arrangement of electrons, they can design a probe with a specific tip geometry and material properties to achieve their goals.

The Potential for Computer-Controlled Chemistry

This breakthrough has significant implications for computer-controlled chemistry. If researchers can control the tip's structure down to the single atomic level, they may be able to manipulate individual chemical bonds or even assemble complex molecules at will. This would revolutionize our ability to design and synthesize new materials and compounds with specific properties.

Calibrating Camera Systems

Another aspect of this technology is equivalent to calibrating camera systems. Researchers need to ensure that the probe's imaging characteristics are matched to the object being imaged, in order to produce high-quality images. This involves adjusting the probe's radius of curvature, sharpness, and material properties to optimize its performance.

The Role of Symmetry

Interestingly, symmetry plays a crucial role in this process. The shape of the probe is critical in determining how it interacts with the surface being imaged. A blunt tip, such as a double tip or triple tip, may produce images that are symmetrically equivalent to those produced by a sharp tip, but with important differences. By understanding these symmetry relationships, researchers can design probes and imaging protocols that take advantage of these effects.

The Test Structure

A key part of this technology is the development of test structures that allow researchers to map out the probe's performance. These structures are designed to reveal the tip's characteristics, such as its shape, material properties, and optical interactions with the surface. By using these test structures, researchers can optimize their probes and imaging protocols to achieve high-quality images.

Controlled Experiments

Another innovative aspect of this technology is the use of controlled experiments to expose specific regions of the sample surface to light. This allows researchers to deliberately create areas where the polymer becomes soluble, which can be used to manipulate individual chemical bonds or assemble complex molecules. By controlling these conditions, researchers can now explore new possibilities for computer-controlled chemistry.

The Future of Research

The development of this technology has significant implications for research in fields such as materials science, chemistry, and biology. With the ability to control probes at the single atomic level, researchers may be able to design and synthesize new materials and compounds with specific properties. This could lead to breakthroughs in fields such as energy storage, catalysis, and medicine. As this technology continues to evolve, we can expect to see significant advances in our understanding of matter at the atomic level.

"WEBVTTKind: captionsLanguage: enyou've got physics-y things in front of you i've got physically things in front of me i've got also got a paper i'm hugely hugely excited about um unfortunately not from our group i really wish i could say it was from our group we've been interested in this type of thing for for many years so as you can see it's this autonomous scanning probe microscopy in-situ tip conditioning through machine learning and this is from bob walker's group uh in the university of alberta and it's a phenomenal piece of work um why i think it might be of interest to computer file audience it could also be of interest to 60 symbols audience but for this in particular it's they're basically controlling the atomic structure of matter through machine learning so it's really the interface between physics and computer science first of all it's all about scanning probe microscopy now i think we've talked a little bit about scanning for microscopy but in essence what we have is a sharp tip preferably atomically sharp with one atom sticking out the end and that sounds really difficult to do but in practice not that difficult to do what is difficult to control is just how many atoms stick out the end and the arrangement of the atoms at the end and that is really the bane of any scanning probe microscopic's life but let's say we've managed to create this we've got one atom sticking at the end here's our surface here's our sample you bring it in and you move it so you're almost at the touching point sometimes even at the contact point whereby you have a very small separation between the the tip and the sample so what we have is our sharp tip we bring it in really close to the surface and we do that using piezoelectric crystals actually those barbecue lighter things that you use where you click at the end pitot electric crystals in those so you bring it in really close within an atomic diameter or so and then you move it across the surface and what you're picking up on is the interaction of that atom right at the end of the tip with the atoms at the surface so you move back and forth back and forth and build up an image that way it's a slow process but it's a process that is incredibly precise and we sacrifice speed for that precision and the other great thing is because it's a probe as well as imaging what you can do is you can deliberately try to manipulate the surface so you can bring the tip in bond and try to pluck atoms out for example or if they're atoms absorbed in the surface you can try and slide them move them across and we're now at the point where in many cases we can slide atoms of will across the surface to spell out different patterns but but and there's a big but as i said the key thing here with scanning for microscopy is controlling the structure at the end of the probe and so we'll start off we will take our tip and the way we create our tip the first stage in creating the tip is we put it into a solution solution of sodium hydroxide i know this is chemistry of the computer file audience bear with me you etch it down to a sharp point but that's not good enough generally it's not good enough then you put it in your vacuum system you move it in you'd like to be able to see atoms but most the time you don't see atoms so what do you do well you apply a little voltage pulse to try and jiggle the atoms around at the end of the probe or you apply an increase in the current and that can lead to effects which will move the atoms around at the end of the probe that doesn't work you crush it gently in that doesn't work you crush it a little bit further in that doesn't work you drive it in half a millimeter or so and push it around and try to try to jiggle it around how do you know if it's worked very very good question you go does this image look good really yeah what we really want to have is as an image whereby what we're seeing is the structure of the surface and that this has as little influence as possible so what we really want to do is have a sharper probe as possible because for example if we have a probe that looks like this where we've got a flat plane where each one of these atoms will could potentially contribute to the image that can lead to very confusing images because you've got multiple different imaging centers let's say we have something like this so it comes in like this to the surface where you've got an atom here and an atom here and both of those can contribute to the image and sometimes it'll be tilted slightly and this one can contribute more because it's closer to the surface but this one can still contribute it's reasonably close to the surface so those are double tip images they're the bane of the scanning pro microscope is life because really what you want to do is to get right down to that atomic level at the moment what happens is students post docs researchers sit in the lab and they they literally just trial and error push the tip into the surface apply voltage pulses try and coerce it into the state they want and it's a massive bottleneck massive massive bottleneck when in fact what we'd want is a sort of auto tune or an autofocus button where you get to the end of the day if you've experienced your experiments might have worked okay but perhaps your tip is not in a good state and you want to press a button which is optimized probe and go home and then come in the morning and do your proper experiments instead of driving yourself to distraction just pushing the tip into the surface to try and change the structure of the tip and that's been a long time coming scanning for microscopy has been around since the early 80s and we're still at that level many and nottingham's no different many many groups are still at that level where it's a you know a phd student in their second year literally banging my head against the table going please work that's why this paper is so important and it's it's really nice so what they've done they've trained a convolutional neural net and there are many many good videos about neural nets and convolutional neural nets and computer file mike pound has done a number of really good videos i thoroughly recommend them that's what they've used they've managed to train it to distinguish between when it's got a double tip and when it's got a proper single atom tip and they've gone that extra step and they've modified the probe so that it will recover the structure you need to get a good single atom image as it were this is a really big leap forward and for me it's it's really exciting i'm slightly disappointed that we didn't get that we've been working on this type of problem for for a number of years actually another good reason why um it's good to do computerified video and there's actually with computer scientists here in nottingham and a number of years back we tried to employ what are called evolutionary algorithms or genetic algorithms to try to tune the probe that way it worked okay it didn't work particularly well all the time and the way to go obviously the next step the next evolutionary stage in that was to go to the machine learning side of things and i actually had a summer intern working on this last year but we just as i said we were beaten to pip to the post the great thing is is where it can go because they've controlled the probe and they've been going for a particular atomic resolution but actually sometimes particularly if you want to do chemistry let's use this one actually sometimes that atom sticking out the end is what you need but sometimes what you might actually need just in terms of the chemistry in the arrangement of the electrons if you really want to manipulate atoms if you want to do chemistry computer controlled chemistry you might actually need a structure a bit more like this in terms of hybrid bonds and how it interacts with the surface the next step in this is to not just give us a good tip give us a tip with our particular structure give us a tip with a particular state and then we are really not just controlling matter at the atomic level we're controlling it right down to the single chemical bond level and in fact we're controlling the right the quantum mechanical structure of the tip when we go a little bit further along these lines you know can we actually just tell it we wanted to build something like that we just wanted you know can we get the the computer to to as long as that's within the laws of physics that's our blueprint you know and that might be a chunk of silicon can we actually get it to build that if you're calibrating a camera there are various test images that you can use to to calibrate it from be it white balance or focus or whatever is there an equivalence in this that's effectively what they do when they talk about a single tip versus a double tip this is a probe this is a sample so what we would like is that the radius of curvature the sharpness of this is smaller than the object we're actually imaging at the surface so if we do that as we trace it across we'll map out an image of what's happening at the surface however that's just by symmetry exactly equivalent to that so if you have a relatively blunt tip like a double tip or like a triple tip or a quadruple tip or a cluster and you've got something sharp at the surface like a single atom or a single bond that sticks out of the surface then what will happen is that this will image this so this is our test structure to come into to to map it across to the video uh process this is our test structure this is allowing us to see what's happening at the end of the probe and that's exactly what they're doing in this they have single bonds which are sticking out the surface when they see these sort of ghost-like images it's telling us about the structure the tip rather than the structure of the surface and then they control it accordingly so this is our control it's built into the experiment and you expose the plastic at the surface in particular regions according to the pattern and the important thing is when the this particular polymer is exposed to light it becomes soluble the regionyou've got physics-y things in front of you i've got physically things in front of me i've got also got a paper i'm hugely hugely excited about um unfortunately not from our group i really wish i could say it was from our group we've been interested in this type of thing for for many years so as you can see it's this autonomous scanning probe microscopy in-situ tip conditioning through machine learning and this is from bob walker's group uh in the university of alberta and it's a phenomenal piece of work um why i think it might be of interest to computer file audience it could also be of interest to 60 symbols audience but for this in particular it's they're basically controlling the atomic structure of matter through machine learning so it's really the interface between physics and computer science first of all it's all about scanning probe microscopy now i think we've talked a little bit about scanning for microscopy but in essence what we have is a sharp tip preferably atomically sharp with one atom sticking out the end and that sounds really difficult to do but in practice not that difficult to do what is difficult to control is just how many atoms stick out the end and the arrangement of the atoms at the end and that is really the bane of any scanning probe microscopic's life but let's say we've managed to create this we've got one atom sticking at the end here's our surface here's our sample you bring it in and you move it so you're almost at the touching point sometimes even at the contact point whereby you have a very small separation between the the tip and the sample so what we have is our sharp tip we bring it in really close to the surface and we do that using piezoelectric crystals actually those barbecue lighter things that you use where you click at the end pitot electric crystals in those so you bring it in really close within an atomic diameter or so and then you move it across the surface and what you're picking up on is the interaction of that atom right at the end of the tip with the atoms at the surface so you move back and forth back and forth and build up an image that way it's a slow process but it's a process that is incredibly precise and we sacrifice speed for that precision and the other great thing is because it's a probe as well as imaging what you can do is you can deliberately try to manipulate the surface so you can bring the tip in bond and try to pluck atoms out for example or if they're atoms absorbed in the surface you can try and slide them move them across and we're now at the point where in many cases we can slide atoms of will across the surface to spell out different patterns but but and there's a big but as i said the key thing here with scanning for microscopy is controlling the structure at the end of the probe and so we'll start off we will take our tip and the way we create our tip the first stage in creating the tip is we put it into a solution solution of sodium hydroxide i know this is chemistry of the computer file audience bear with me you etch it down to a sharp point but that's not good enough generally it's not good enough then you put it in your vacuum system you move it in you'd like to be able to see atoms but most the time you don't see atoms so what do you do well you apply a little voltage pulse to try and jiggle the atoms around at the end of the probe or you apply an increase in the current and that can lead to effects which will move the atoms around at the end of the probe that doesn't work you crush it gently in that doesn't work you crush it a little bit further in that doesn't work you drive it in half a millimeter or so and push it around and try to try to jiggle it around how do you know if it's worked very very good question you go does this image look good really yeah what we really want to have is as an image whereby what we're seeing is the structure of the surface and that this has as little influence as possible so what we really want to do is have a sharper probe as possible because for example if we have a probe that looks like this where we've got a flat plane where each one of these atoms will could potentially contribute to the image that can lead to very confusing images because you've got multiple different imaging centers let's say we have something like this so it comes in like this to the surface where you've got an atom here and an atom here and both of those can contribute to the image and sometimes it'll be tilted slightly and this one can contribute more because it's closer to the surface but this one can still contribute it's reasonably close to the surface so those are double tip images they're the bane of the scanning pro microscope is life because really what you want to do is to get right down to that atomic level at the moment what happens is students post docs researchers sit in the lab and they they literally just trial and error push the tip into the surface apply voltage pulses try and coerce it into the state they want and it's a massive bottleneck massive massive bottleneck when in fact what we'd want is a sort of auto tune or an autofocus button where you get to the end of the day if you've experienced your experiments might have worked okay but perhaps your tip is not in a good state and you want to press a button which is optimized probe and go home and then come in the morning and do your proper experiments instead of driving yourself to distraction just pushing the tip into the surface to try and change the structure of the tip and that's been a long time coming scanning for microscopy has been around since the early 80s and we're still at that level many and nottingham's no different many many groups are still at that level where it's a you know a phd student in their second year literally banging my head against the table going please work that's why this paper is so important and it's it's really nice so what they've done they've trained a convolutional neural net and there are many many good videos about neural nets and convolutional neural nets and computer file mike pound has done a number of really good videos i thoroughly recommend them that's what they've used they've managed to train it to distinguish between when it's got a double tip and when it's got a proper single atom tip and they've gone that extra step and they've modified the probe so that it will recover the structure you need to get a good single atom image as it were this is a really big leap forward and for me it's it's really exciting i'm slightly disappointed that we didn't get that we've been working on this type of problem for for a number of years actually another good reason why um it's good to do computerified video and there's actually with computer scientists here in nottingham and a number of years back we tried to employ what are called evolutionary algorithms or genetic algorithms to try to tune the probe that way it worked okay it didn't work particularly well all the time and the way to go obviously the next step the next evolutionary stage in that was to go to the machine learning side of things and i actually had a summer intern working on this last year but we just as i said we were beaten to pip to the post the great thing is is where it can go because they've controlled the probe and they've been going for a particular atomic resolution but actually sometimes particularly if you want to do chemistry let's use this one actually sometimes that atom sticking out the end is what you need but sometimes what you might actually need just in terms of the chemistry in the arrangement of the electrons if you really want to manipulate atoms if you want to do chemistry computer controlled chemistry you might actually need a structure a bit more like this in terms of hybrid bonds and how it interacts with the surface the next step in this is to not just give us a good tip give us a tip with our particular structure give us a tip with a particular state and then we are really not just controlling matter at the atomic level we're controlling it right down to the single chemical bond level and in fact we're controlling the right the quantum mechanical structure of the tip when we go a little bit further along these lines you know can we actually just tell it we wanted to build something like that we just wanted you know can we get the the computer to to as long as that's within the laws of physics that's our blueprint you know and that might be a chunk of silicon can we actually get it to build that if you're calibrating a camera there are various test images that you can use to to calibrate it from be it white balance or focus or whatever is there an equivalence in this that's effectively what they do when they talk about a single tip versus a double tip this is a probe this is a sample so what we would like is that the radius of curvature the sharpness of this is smaller than the object we're actually imaging at the surface so if we do that as we trace it across we'll map out an image of what's happening at the surface however that's just by symmetry exactly equivalent to that so if you have a relatively blunt tip like a double tip or like a triple tip or a quadruple tip or a cluster and you've got something sharp at the surface like a single atom or a single bond that sticks out of the surface then what will happen is that this will image this so this is our test structure to come into to to map it across to the video uh process this is our test structure this is allowing us to see what's happening at the end of the probe and that's exactly what they're doing in this they have single bonds which are sticking out the surface when they see these sort of ghost-like images it's telling us about the structure the tip rather than the structure of the surface and then they control it accordingly so this is our control it's built into the experiment and you expose the plastic at the surface in particular regions according to the pattern and the important thing is when the this particular polymer is exposed to light it becomes soluble the region\n"