WEBVTT 00:13:49.000 --> 00:13:57.000 I'm not sure if all is connected, it doesn't seem…. Uh, maybe…. 00:13:57.000 --> 00:13:58.000 Oh, okay. Maybe, uh, to step out. 00:13:58.000 --> 00:14:05.000 He was here, like, a few minutes ago. Uh, I think that, uh, Christian, like, Christian went to check the other link. They were all…. 00:14:05.000 --> 00:14:06.000 Now, some people, uh, in the other…. Oh, yeah, cruising your back. 00:14:06.000 --> 00:14:15.000 Yeah, yeah. Yeah, Jeff… Jeff Nelson went out to, uh. 00:14:15.000 --> 00:14:18.000 Tell the other attendees, you should be right back. 00:14:18.000 --> 00:14:24.000 Okay. Is Paul over there? 00:14:24.000 --> 00:14:25.000 Rewind some tests. 00:14:25.000 --> 00:14:42.000 Once we have multiple. 00:14:42.000 --> 00:14:49.000 And that is bizarre. We have two meetings. 00:14:49.000 --> 00:15:19.000 Yeah, this one, the same length from last week and the previous meetings, and then there's another… there's a new one for today for some reason. 00:15:26.000 --> 00:15:31.000 Hang on, maybe we should just start. 00:15:31.000 --> 00:15:34.000 I was going to wait to see if Jeff…. 00:15:34.000 --> 00:15:35.000 That also…. Maybe we could stop. 00:15:35.000 --> 00:15:39.000 Okay. 00:15:39.000 --> 00:15:43.000 Yeah, why don't you go ahead and start? 00:15:43.000 --> 00:15:53.000 Okay, that's fine, thank you. So, yeah, good afternoon, everyone. My name is Grant Johnson. I'm a scientist at Pacific Northwest National Lab. 00:15:53.000 --> 00:15:58.000 Um, we're still getting some noise there, so if you could mute yourself, please, that'd be great. 00:15:58.000 --> 00:16:05.000 So, um, pleased today to give you an overview of our microelectronics Science Research Center project. 00:16:05.000 --> 00:16:11.000 Uh, which is titled, Assembly of Molecular Memoristers for Energy-Efficient Computing. 00:16:11.000 --> 00:16:18.000 And the sort of theme of the overall project is captured by this graphics that we just recently had made up. 00:16:18.000 --> 00:16:24.000 Uh, where we have molecular memoristers, in this case, which are actually molecular metal oxides, or polyoxymetolates. 00:16:24.000 --> 00:16:30.000 And then we use, uh, biological scaffolds, in this case peptides, to arrange those into. 00:16:30.000 --> 00:16:41.000 Device-level structures for testing, so…. I'll give kind of an overview of the project, and I'll talk about my thrust, which is focused on sort of local interactions. 00:16:41.000 --> 00:16:46.000 I'll then hand it off to Chen Long Chen, who's the project manager for this project here at PNNL. 00:16:46.000 --> 00:17:01.000 And he'll tell you more about the macromolecular molecular assembly. And then Zhao Gun Lang will tell you about the device testing that we're doing with these materials. 00:17:01.000 --> 00:17:14.000 So, uh, I think this audience pretty would, uh, knows pretty well, uh, the goals of neuromorphic computing, uh, and the, uh, desire to build biologically-inspired systems that mimic the brain's parallel event-driven. 00:17:14.000 --> 00:17:27.000 And energy-efficient information processing. Uh, so this is inherently different to, sort of, von Neumann computing architectures, where you're shuffling information back and forth between computers and memory. 00:17:27.000 --> 00:17:40.000 And we need materials that basically function as biological synapses. Uh, so there's a lot of work that's been done on neuromorphic chips and architectures like these memoristic crossbar arrays. 00:17:40.000 --> 00:17:46.000 But there are still some material challenges associated with these, and we think that our molecular approach. 00:17:46.000 --> 00:17:51.000 Two memristers might have some inherent advantages, uh, to address some of these issues. 00:17:51.000 --> 00:18:02.000 So, um, Minristers play a key role, uh, in neuromorphic computing. Uh, this is due to their nonlinear dynamics, their multi-level or analog conductance. 00:18:02.000 --> 00:18:06.000 And the fact that they can be a very low power use in certain implementations. 00:18:06.000 --> 00:18:20.000 Uh, they also have the advantages of being non-volatile, fast. And again, allow… overcoming that von Benoemann bottleneck. So, the traditional architecture usually looks something like this. You have your enristive material. 00:18:20.000 --> 00:18:27.000 With a top and a bottom electrode. And then the characteristics of this, uh, the voltage current response. 00:18:27.000 --> 00:18:36.000 Looks something like this for a digital response, and like this for an analog response, which is more what we're interested in for these neuromorphic computing applications. 00:18:36.000 --> 00:18:45.000 And the market for enristers is predicted to grow dramatically over the coming years, as these things are increasingly used in IT telecoms. 00:18:45.000 --> 00:18:56.000 Uh, and in other industries. So, as I mentioned, um, despite, you know, all their advantages, there are also some challenges that need to be overcome. 00:18:56.000 --> 00:19:02.000 Uh, one of these, uh, clearly, is the variability, the device-to device, and cycle-to-cycle. 00:19:02.000 --> 00:19:15.000 Variability. In addition, degradation. And both of these basically relate to the stochastic switch switching mechanisms that impede reproducible memorister performance. 00:19:15.000 --> 00:19:21.000 So, on the right-hand side there are just the various mechanisms by which these materials can function. 00:19:21.000 --> 00:19:26.000 Uh, in our case, we're concerned with metal oxide, so the two most important. 00:19:26.000 --> 00:19:32.000 Phenomena, of course, are oxygen vacancy, migration, and also conductive filament formation. 00:19:32.000 --> 00:19:37.000 And we'll explain a little bit about how our molecular memoristers can be used to modulate both. 00:19:37.000 --> 00:19:47.000 Uh, of those effects. So, uh, this is kind of a overview figure that summarizes the objectives of our, um, project. 00:19:47.000 --> 00:19:55.000 Uh, we aim to self-assemble molecular membristors with long-range order for resilient and energy-efficient neuromorphic computing. 00:19:55.000 --> 00:20:01.000 Um, so you can see there on the left-hand side, uh, hopefully you can see my laser pointer. 00:20:01.000 --> 00:20:06.000 Um, we want to take these molecular metal oxides, which are sort of nanometer scale. 00:20:06.000 --> 00:20:18.000 Molecular metal oxides. These have multiple, uh, electronic charge states, or redox states, each of which gives a discrete current voltage response. 00:20:18.000 --> 00:20:25.000 But we need a way to put those into a material with long-range order that's well-organized on a 2D electrode interface. 00:20:25.000 --> 00:20:31.000 So, to do that, we're leveraging peptides, which are biomolecular polymers. 00:20:31.000 --> 00:20:36.000 Which will allow us to assemble these polyoxylmetolates into more ordered structures. 00:20:36.000 --> 00:20:44.000 Creating individual synapses, basically, for a single polyoxomethylate. And then into larger, uh, synaptic networks. 00:20:44.000 --> 00:20:55.000 So, the key games here are controlling that local environment around this molecular metal oxide to give it the current resistance performance that we want. 00:20:55.000 --> 00:21:01.000 And then arranging it into longer-range order on electrode materials that are compatible. 00:21:01.000 --> 00:21:11.000 Uh, with conventional CMOS technology technologies. So the general workflow on the right looks something like this, where we design and control these memoristive states locally. 00:21:11.000 --> 00:21:19.000 We tune this at a larger scale, and then we start going up to signal and memory functionality, and then evaluating these devices for energy efficiency. 00:21:19.000 --> 00:21:30.000 And, uh, neuromorphic computing applications. Our project addresses three scientific knowledge gaps, uh, through three integrated research thrusts. 00:21:30.000 --> 00:21:38.000 Uh, the first one that's led by myself is focused on the local interactions between these polyoxomethylates and sequence-defined peptides. 00:21:38.000 --> 00:21:44.000 And what effect that has on their memoristic states and absorption behavior on 2D interfaces. 00:21:44.000 --> 00:21:51.000 The second knowledge gap is centered around the effects of peptide sidechain chemistry, structure, and composition. 00:21:51.000 --> 00:21:56.000 And how that can be tuned to control the long-range order of these assemblies. 00:21:56.000 --> 00:22:03.000 And also the formation of these nano-confined ion transport channels through which these conductive filaments will form. 00:22:03.000 --> 00:22:08.000 And the third thrust, uh, led by Xiaogen Lang from University of Michigan. 00:22:08.000 --> 00:22:14.000 Is focused on assembling these materials into devices and actually testing their energy efficiency. 00:22:14.000 --> 00:22:24.000 For computing operations. So, uh, here are some images that capture the thrust that I described on the previous slide. 00:22:24.000 --> 00:22:31.000 Um, we actually take a slightly different approach to many of the people in this center in that we're using solution-phase synthesis. 00:22:31.000 --> 00:22:35.000 To make these polyoxometallates, which are shown on the left-hand side here. 00:22:35.000 --> 00:22:41.000 So we can tune what sort of transition metals are in these, the size and charge states of these metal oxides. 00:22:41.000 --> 00:22:49.000 And then we also connect them, uh, through different covalent linkages, uh, to polyoxometal… or to, uh, peptides, which serve. 00:22:49.000 --> 00:22:55.000 Um, as the host matrix to create an ordered array. So we can do, um…. 00:22:55.000 --> 00:23:01.000 Deposition experiments, where we create these in very clean environments on 2D interfaces, and we interrogate. 00:23:01.000 --> 00:23:07.000 The current voltage response of individual clusters, uh, using scanning tunneling microscopy. 00:23:07.000 --> 00:23:19.000 And infrared spectroscopy. But then we start to assemble these with peptides into larger-scale arrays, which we can then characterize using larger-scale techniques, such as conductive AFM. 00:23:19.000 --> 00:23:31.000 Uh, and simulations such as molecular dynamics. The third thrust that involves taking these materials and putting them into both lateral and vertical memorister configurations. 00:23:31.000 --> 00:23:36.000 And analyzing their current voltage characteristics, the memristive stays, and evaluating them. 00:23:36.000 --> 00:23:42.000 For computing applications. 00:23:42.000 --> 00:23:49.000 So, uh, this is just a picture of the team. So this project is led out of Pacific Northwest National Laboratory. 00:23:49.000 --> 00:23:54.000 I serve as the lead for Thrust One. Chen Long Chen leaves the peptide task. 00:23:54.000 --> 00:24:00.000 And then the folks shown on the bottom here bring a whole range of expertise in material synthesis. 00:24:00.000 --> 00:24:10.000 Controlled deposition, multimodal characterization, and modeling. And then we have some postdocs, which are going to be joining us shortly at the end of the July and August to push forward. 00:24:10.000 --> 00:24:20.000 This effort in the laboratory. Uh, we've partnered, uh, with several other institutes, uh, shown here. 00:24:20.000 --> 00:24:26.000 Uh, to bring in some additional expertise and theoretical modeling, machine learning, and device fabrication and testing. 00:24:26.000 --> 00:24:31.000 So we have Tin Cow from University of Washington is helping us with our surface simulations. 00:24:31.000 --> 00:24:36.000 Xiaogen, who I mentioned, is leading our device assembly and testing thrust. 00:24:36.000 --> 00:24:45.000 And then go away, who brings expertise in machine learning models to help us explore the broad parameter space of these materials and select those that are most promising. 00:24:45.000 --> 00:24:56.000 Uh, for our applications. And then Jan at Berkeley is also helping us with the deposition and device testing. 00:24:56.000 --> 00:25:04.000 So, our first thrust, uh, again, that I lead is focused on the local interactions in the immediate region around. 00:25:04.000 --> 00:25:18.000 These molecular metal oxides. So, as you can see here, uh, these clusters are on the order of a half nanometer in size. Uh, they have different heteroatoms at the center, things like phosphorus, silicon, and germanium, which can be used to tune. 00:25:18.000 --> 00:25:30.000 The, uh, energy levels of the electronic orbitals. You can also exert a lot of control over those orbitals through the substitution of the transition metals, things like vanadium, molybdenum, and tungsten. 00:25:30.000 --> 00:25:40.000 So we will be exploring that through selective substitution reactions. The next step, of course, is then covalently attaching these molecular metal oxides to. 00:25:40.000 --> 00:25:45.000 Uh, the polymer, uh, the bio… the biopolymer, the peptide in this case. 00:25:45.000 --> 00:25:52.000 And we're exploring both file, click chemistry, and condensation reactions as ways to achieve that. 00:25:52.000 --> 00:25:59.000 We have unique capabilities, which I'll tell you a little bit about in a minute, where we can deposit these sort of complex molecules from the gas phase. 00:25:59.000 --> 00:26:08.000 Using a technique known as ion soft landing. This allows us to create very well-defined interfaces that are well-suited to characterization. 00:26:08.000 --> 00:26:19.000 With spatially resolved techniques, like scanning, tunneling, microscopy, and spectroscopy. So this allows us to get the sort of individual behavior, um, current voltage response. 00:26:19.000 --> 00:26:29.000 The memristive states of these polyoxomatylates on surfaces, and understand how that's tuned by the parameters that I just described. 00:26:29.000 --> 00:26:37.000 So, uh, here's some evidence of what I was just speaking about. So, if you have one of these molecular metal oxides on a surface. 00:26:37.000 --> 00:26:43.000 They can have up to 4 different conductive states, or 4 logic states. 00:26:43.000 --> 00:26:46.000 Uh, this is shown here on the right-hand side, again, in these plots. 00:26:46.000 --> 00:26:51.000 So, each one of those has the potential to be used to store a certain. 00:26:51.000 --> 00:26:59.000 Bit, or to, um, actually provide a certain synaptic weight, uh, in a, uh, neural network. 00:26:59.000 --> 00:27:09.000 In addition to controlling their electronic conductivity, these polyoxylmetolates can also regulate cation migration and conductive filament formation. 00:27:09.000 --> 00:27:19.000 Um, so as you know, one of the big problems with memristors is this stochastic filament formation, which is often not reproducible cycle to cycle or device to device. 00:27:19.000 --> 00:27:31.000 And polyoxymetalates, which have these localized unoccupied molecular orbitals. Can basically serve as transit corridors for these ions, making those processes much more. 00:27:31.000 --> 00:27:41.000 Reproducible. So, uh, one technique that we've developed here at our lab, uh, is this technique known as ion soft landing. 00:27:41.000 --> 00:27:52.000 And the key advantage here is it allows us to take a mixture containing many complex things, so this is a mass spectrum of these molecular metal oxides. 00:27:52.000 --> 00:27:58.000 And if you were to drop cast these or try to solution cast them, you would get everything present in solution. 00:27:58.000 --> 00:28:05.000 But with this mass spectrometry approach, we can individually select each of these ions and isolate it on a surface. 00:28:05.000 --> 00:28:13.000 And examine the effect of stoichiometry, heteroatom substitution. Uh, and, uh, covalent functionalization. 00:28:13.000 --> 00:28:24.000 And then we have some in-situ capabilities to actually examine. The structure of these things on the interface, and how they change when we apply a voltage and change the redox or conductive state. 00:28:24.000 --> 00:28:36.000 Of these metal oxides. And we have some capabilities as well for UHV sample transfers, so these can be moved over to STM and STS and characterized without exposure to air. 00:28:36.000 --> 00:28:42.000 So, one of the things we think is going to be quite important for these molecular memoristers. 00:28:42.000 --> 00:28:52.000 Are the local interactions between the polyoxymetalate and the peptide. Uh, with the support, and then also with adjacent molecular metal oxides. 00:28:52.000 --> 00:28:59.000 And one way we can get information about this is by studying the spectroscopy of these things as a function of their surface coverage. 00:28:59.000 --> 00:29:03.000 And this is an example here from a paper from a couple years ago. 00:29:03.000 --> 00:29:14.000 Or we basically imaged these materials on a surface as a function of surface coverage and showed how the vibrational modes of these things evolve as a function of almost 40 wavenumbers. 00:29:14.000 --> 00:29:17.000 From an environment where they're interacting mostly with the electrode surface. 00:29:17.000 --> 00:29:25.000 To an environment where they're interacting with each other in a higher coverage regime. 00:29:25.000 --> 00:29:30.000 As I mentioned, we are going to study the individual minerista properties of these. 00:29:30.000 --> 00:29:45.000 Polyxamethylates and their assemblies on surfaces. And this is… you're done using PNNL's STM and STS capabilities shown here. We have several instruments, both for low temperature and variable temperature measurements. 00:29:45.000 --> 00:29:57.000 And this is also coupled with X-ray photoelectron spectroscopy and the ability to dose reactive gases and examine the effect of those environmental conditions. 00:29:57.000 --> 00:30:03.000 So, the project, uh, just started this year, um, so, uh, we do have some exciting results to share. 00:30:03.000 --> 00:30:11.000 These are some results of these molecular metal oxides deposited onto graphite. These are conductive AFM images shown here. 00:30:11.000 --> 00:30:20.000 At two different coverages. And then we also have current voltage and DIDV curves here, revealing that, in fact, these materials do show multiple. 00:30:20.000 --> 00:30:29.000 Discrete, uh, conductance states, which we believe will be useful. From Memrist or applications. 00:30:29.000 --> 00:30:35.000 So, um, I think this is the last slide for my thrust. Um, I just want to emphasize again that. 00:30:35.000 --> 00:30:45.000 The polyoxometylates themselves are one mineristive element, but then we need a way to arrange those into macromolecular assemblies with long-range order. 00:30:45.000 --> 00:30:52.000 And to do that, we have to covalently attach them to the peptides. So we're exploring various tethering schemes. 00:30:52.000 --> 00:31:01.000 Uh, using this TRIS alkoxyl chemistry shown here. Uh, where we can form both amide bonds and also do click chemistry. 00:31:01.000 --> 00:31:12.000 Uh, to create strong covalent bonds to the host scaffold. So, um, again, I'm just going to bring you back to our high-level diagram. So that, um…. 00:31:12.000 --> 00:31:20.000 Summarizes the sort of work that I'm doing on the local interactions of these memristers. I'm going to hand it off now to my colleague, Chun Long Chen, who's going to tell you more about. 00:31:20.000 --> 00:31:26.000 The macromolecular assembly and peptide work. 00:31:26.000 --> 00:31:33.000 Okay, thanks, Grant. So, as Greg mentioned, also, uh, in…. 00:31:33.000 --> 00:31:43.000 From this OO, uh…. Picture, right? So, this project, uh, what we are really, uh, doing is, like, we want to take this BioInspy concept. 00:31:43.000 --> 00:31:48.000 As you know, like, from the top of the figure, right? So, uh, the…. 00:31:48.000 --> 00:31:52.000 Iron transport in the brain, so it's very important for this, uh. 00:31:52.000 --> 00:31:59.000 Energy-efficient neuromorphic computing. So what we try to do for SROS2, right, is. 00:31:59.000 --> 00:32:04.000 How we can also, uh, create this kind of long-range ordering, right, to…. 00:32:04.000 --> 00:32:12.000 Tune the iron transport or electron transport. And then we can use a lot of information learned from SRAS-1. 00:32:12.000 --> 00:32:17.000 For example, this, uh, tunable…. I'm a Marista state, right? How we can. 00:32:17.000 --> 00:32:23.000 Organize them into a…. Long-range order, so which kind of show in this, uh. 00:32:23.000 --> 00:32:31.000 Cartoon, right? If we develop this, uh, framework structure. And then how we can use these, uh, different charge state. 00:32:31.000 --> 00:32:40.000 And then the ordering of the peptide to legate the. Iron transport, electron transport for building energy-efficient device. 00:32:40.000 --> 00:32:48.000 Uh, Grant, can you… are you controlling the slide? 00:32:48.000 --> 00:33:00.000 Oh, okay, yes, thanks. So, uh…. Yeah, thanks, Grant. Uh, like I mentioned, right, the SWAS II, what we really want to do is now say, okay, how we can. 00:33:00.000 --> 00:33:06.000 Use the self-assembly or the peptor into crystalline material. To organize, uh. 00:33:06.000 --> 00:33:13.000 This, uh, memorized the state of material, right, into this long-range ordering, and then. 00:33:13.000 --> 00:33:19.000 To do that, we need to put this memorized the material onto a substrate, right? 00:33:19.000 --> 00:33:25.000 Which could be the, uh, the memorizedo electrolyte, or this semiconducting material. 00:33:25.000 --> 00:33:31.000 So we originally put four different ways to make this kind of material. 00:33:31.000 --> 00:33:38.000 One is, for example, if we make this self-assembled crystalline material, we can do this surface agnostic coating. 00:33:38.000 --> 00:33:45.000 By doing that, we will have, layer by layer, this ordered material onto the substrate surface. 00:33:45.000 --> 00:33:51.000 To, like, really, the ion or electron, uh. Transport, for example, if we have the. 00:33:51.000 --> 00:34:00.000 So, uh, electronized, the solar ion. Diffusion through the filament, so then we can use peptide and POM to. 00:34:00.000 --> 00:34:08.000 Control that stage. And we also… we used to put, like, a PCVD to do this, uh, PCVD. 00:34:08.000 --> 00:34:19.000 Close of the peptide material… crystalline material on the surface. But then, due to the scope of the work, we kind of, like, removed the…. 00:34:19.000 --> 00:34:24.000 Pcvd from, uh, our…. Task, uh, temporary. 00:34:24.000 --> 00:34:32.000 So, Grant, next slide. Yeah, so, again, so this is kind of, uh, repeat what I…. 00:34:32.000 --> 00:34:37.000 Say, right? Like, you know, this last two, it's really say, okay. 00:34:37.000 --> 00:34:43.000 For stress one, once we understand a lot of this electronic structure and how the. 00:34:43.000 --> 00:34:49.000 Local interaction, which could be. The covalent attachment of the peptide with the palm. 00:34:49.000 --> 00:34:58.000 Or peptide sequence, right, influence the pump charge state. Now the question is, like, how we can use those molecular level understanding. 00:34:58.000 --> 00:35:04.000 Now build it into this crystalline material, right, to tune the electron and ion transport. 00:35:04.000 --> 00:35:14.000 So then, we can deposit this material onto the. Uh, this, uh, memorizedo electrolyte, so, like, to do, uh, to develop this. 00:35:14.000 --> 00:35:22.000 Organic… inorganic interface, right? So, to understand. For example, some of the key questions we want to, uh. 00:35:22.000 --> 00:35:32.000 Understand down the road, it's like. How the peptide sequence, or the interaction in… interact with this, uh, inorganic interface. 00:35:32.000 --> 00:35:37.000 And how we can actually use the inorganic with specific lattice to influence the. 00:35:37.000 --> 00:35:47.000 Ordering of this, uh, peptide pump, uh, structure. And then, by doing that, we can, you know, get a lot of fundamental understanding how the. 00:35:47.000 --> 00:35:52.000 Uh, Iron, or the electron. Flow during the Memorista, uh…. 00:35:52.000 --> 00:35:59.000 Device, uh…. The performance process. And next slide, Grant. 00:35:59.000 --> 00:36:06.000 So, this is basically, uh, one slice to show. How, uh, what we had done before. 00:36:06.000 --> 00:36:13.000 We can actually create a crystalline material, right? So the, uh, left figure showing. 00:36:13.000 --> 00:36:20.000 If we keep the same hydrophobic domain. So we can add a lot of different functional groups. 00:36:20.000 --> 00:36:29.000 To do this, basically, kind of, like, co-criticization process. So then we are able to not only get this crystalline nanomaterial, right, so in…. 00:36:29.000 --> 00:36:34.000 Uh, for this project, we want to. Devote, like, poem containing crystalline material. 00:36:34.000 --> 00:36:43.000 But because of this, uh…. High tunability coming from the peptide, and also coming from this co-crystallization process. 00:36:43.000 --> 00:36:48.000 So, we are able to…. Tune the POM Local Environment. 00:36:48.000 --> 00:36:56.000 Which, in this last one, right, we, uh, Grant mentioned, like, the local interaction is very important to tune in that, uh. 00:36:56.000 --> 00:37:04.000 So there you can imagine, for this thrust, once we align those poems in an ordered way. 00:37:04.000 --> 00:37:09.000 And then we can also tune in the local environment. Basically to achieve some…. 00:37:09.000 --> 00:37:16.000 Kind of like a protein, like…. You know, nature system, right? So the local environment is important for. 00:37:16.000 --> 00:37:20.000 I and electron transport. So the middle figure showing. 00:37:20.000 --> 00:37:31.000 This is actually the Pepto assembly did on the MOS2. The reason we want to use this is because MOS2 is also one of the. 00:37:31.000 --> 00:37:41.000 Very interesting memories, the material, which, uh. Can, you know, tune their bandgap, right, for building their memory device. Here, what we want to do is. 00:37:41.000 --> 00:37:48.000 We want to use this MOS to kind of temporary the peptide assembly, right? So then we are able to. 00:37:48.000 --> 00:37:55.000 Develop this, uh. Heptoid poem and the MOS2 hybrid material use this as a. 00:37:55.000 --> 00:38:06.000 Tunable memorized the material for device. And the right figure is to show, actually, if we understand the lattice of the inorganic material. 00:38:06.000 --> 00:38:20.000 We are able to use this, uh, inorganic lattice, right, which could be HOPG, like conducting material, or in this middle one, like MOS2 semiconductor material. So then we are able to use this. 00:38:20.000 --> 00:38:30.000 Inorganic material. Which could be used as memorized electrolyte, or the buffer material, right? So now, say, okay. 00:38:30.000 --> 00:38:40.000 Use those inorganic lattice to temper the. Ordering of this, uh, peptide poem material, which, uh, obtained from, uh, source 1. 00:38:40.000 --> 00:38:44.000 Next one. Yeah, so this is the…. 00:38:44.000 --> 00:38:58.000 Basically, the result we had before to show. Uh, through the self-assembly process, it is, you know, totally doable. We can, uh, align, uh, like, we can align this, uh. 00:38:58.000 --> 00:39:05.000 Cluster, right? In this case, we use a POM cluster to show if we have the hydrophobic domain. 00:39:05.000 --> 00:39:11.000 Uh, for the tube formation, which is on the…. Left the figure to show. 00:39:11.000 --> 00:39:16.000 And then we can put the palm over there, and then to align this. 00:39:16.000 --> 00:39:27.000 Into nanotube structure, right, which we have well-aligned this nanocluster. So, in this project, we basically used the same approach to align the, uh. 00:39:27.000 --> 00:39:32.000 Different POM cluster. So the middle one to show, if we change hydrophobic domain. 00:39:32.000 --> 00:39:40.000 And then we can get these nanosheath. And then the right image basically also show, like, we can also put this, uh. 00:39:40.000 --> 00:39:47.000 Cluster in the middle to get this highly crystalline material. So this is basically the slides to show. 00:39:47.000 --> 00:39:56.000 Uh, through our previous work. We have demonstrated that align the poem into an ordered way to achieve the long-range order is. 00:39:56.000 --> 00:40:10.000 Totally doable. Next slide. Yeah, so, uh, in this, uh, SWAT, like I mentioned, like, uh, we actually originally put four different ways to make a…. 00:40:10.000 --> 00:40:22.000 This peptidipone material onto the inorganic substrate. So the first one is doing this solution crystallization, which, basically, we started with amorpher, right, then let the peptide and the. 00:40:22.000 --> 00:40:28.000 A pump to crystallize, and then use hydrophobia interaction control the crystallization. 00:40:28.000 --> 00:40:33.000 In the solution, so then we can get a lot of these freestanding crystalline. 00:40:33.000 --> 00:40:40.000 At NanoSheets. The second approach is this temporary approach, which we can use inorganic substrate as the temporary. 00:40:40.000 --> 00:40:50.000 The advantage for this second approach is. Uh, since, uh, yeah, Shogun later going to present Strast 3 to show. 00:40:50.000 --> 00:41:06.000 Uh, they already built some device with MOS2. And then you can imagine, like, once we have this inorganic memorized divide, and then we can actually use this inorganic surface as the temporary further to grow this, uh. 00:41:06.000 --> 00:41:11.000 Pept to a palm, and then use, uh. Compared to a point to further control their. 00:41:11.000 --> 00:41:20.000 The IR and the electron transport. And the third one is about building a Christian framework structure. So here, we kind of remove the…. 00:41:20.000 --> 00:41:27.000 Uh, the PCVD, uh…. Admit that. And then, uh, we will use molecular dynamical simulation to. 00:41:27.000 --> 00:41:35.000 Guide us the material synthesis, and also, uh. Go away and don't change from, uh. 00:41:35.000 --> 00:41:39.000 Michigan State, they are using machine learning to kind of, like, screen. 00:41:39.000 --> 00:41:48.000 They're kinda existing, like, palm crystal. And to see how we can understand more what kind of interaction, what kind of ordering will be important. 00:41:48.000 --> 00:41:54.000 To build the, uh…. The laggerated ion and electron transport. 00:41:54.000 --> 00:42:00.000 And then we also have the in-situ AFM, some other categorization to. 00:42:00.000 --> 00:42:14.000 Look and say, okay, once we build that memorized device. And then what kind of surface charge, and then, like, a, you know, surface potential, or, you know, the bandgap change of the inorganic material will be useful. 00:42:14.000 --> 00:42:23.000 For us to get a certain, uh. Performance from the memorister. For example, Xiao Ganpo will mention, like. 00:42:23.000 --> 00:42:28.000 You know, long-term memory, short-term memory, how we can understand more about the. 00:42:28.000 --> 00:42:34.000 You know, material, uh, property related to those, uh, performance of the memorization. 00:42:34.000 --> 00:42:44.000 Next slide, please. Yeah, so this is one slice to show, for example, right, once we can build a peptide nanosheet. 00:42:44.000 --> 00:42:50.000 So we have this cartoon, which we had a paper last year to show, like, once we have this. 00:42:50.000 --> 00:42:58.000 Highly Chris and nanoSheets. Now we want to build the, uh, functional, right, into this nanosheet. One way we can do is. 00:42:58.000 --> 00:43:02.000 You can imagine those kind of balls, like, right? If those are the poems. 00:43:02.000 --> 00:43:08.000 And now we can use a lot of different chemistry. For example, like those, uh, click chemistry on the right. 00:43:08.000 --> 00:43:17.000 To covertly attach the palm into this, uh, sheets, uh, forming peptide. Next slide, please. 00:43:17.000 --> 00:43:26.000 Yeah, so this is…. The, uh, result to show, once we started with, uh, I think here we have 3 different, uh. 00:43:26.000 --> 00:43:31.000 Palm Cruster, right? So that's because they have different chemistry we can use. 00:43:31.000 --> 00:43:40.000 For, you know, Creek Chemistry, or for, uh. This, uh… I'm a reaction, so then, basically, once we get this, uh. 00:43:40.000 --> 00:43:49.000 Palm containing peptide, and then we are able to use this for self-assembly, and…. 00:43:49.000 --> 00:43:54.000 I think…. Can you guys see my slides? 00:43:54.000 --> 00:43:55.000 I couldn't see it. 00:43:55.000 --> 00:43:58.000 Oh, okay, I think then maybe it just…. You know, because someone mentioned you cannot see the slides. 00:43:58.000 --> 00:44:01.000 It was a mistake. 00:44:01.000 --> 00:44:10.000 Okay, so that would just continue, yeah. So basically, this is to show, like, okay, once we have successful covalent attachment, right, to this, uh. 00:44:10.000 --> 00:44:16.000 Sheets forming peptide, so then we use AFM and SEM to show, indeed, that we have. 00:44:16.000 --> 00:44:21.000 Very nice sheets here. And then, next slide, please. 00:44:21.000 --> 00:44:28.000 And then we also have, uh, the TM image to show indeed, right, we can see individual. 00:44:28.000 --> 00:44:36.000 Like, palm cluster around these, uh, nanosheets. Next slide. 00:44:36.000 --> 00:44:43.000 Yeah, I think now, like, I was, uh…. Transfer this to, uh, Shaogang to tell us how. 00:44:43.000 --> 00:44:46.000 You know, we are doing for the device for Star 3. 00:44:46.000 --> 00:44:54.000 Sanctuin Loan and Grant. And I'll talk about the Thrust 3, and so in this direction, we push on the device Felix. 00:44:54.000 --> 00:45:05.000 And it's about a memory-restricted device fabrication and characterization. For neuromorphic computing. So, as Grand and Chen Long mentioned advantages using. 00:45:05.000 --> 00:45:11.000 The PEP2I and base material to build the memoristers. So here, I want to further emphasize. 00:45:11.000 --> 00:45:17.000 And such a device-oriented effort is driven by this hypothesis that. 00:45:17.000 --> 00:45:23.000 The long range order in the peptide nanostructures, and also there's a wider range of, uh, tunability. 00:45:23.000 --> 00:45:32.000 Of such a macromolecular assemblies. So, can provide the new method for us to create, we call this a biorealistic. 00:45:32.000 --> 00:45:44.000 Synaptic devices. So, for example, and using the different PEP2 nanostructures with different molecular structure, we could build the memory serves. 00:45:44.000 --> 00:45:50.000 With lateral or vertical channel configurations. So they can be programmed. 00:45:50.000 --> 00:45:58.000 To have a short-term or a long-term memory behaviors? What are the different dynamic and the response. 00:45:58.000 --> 00:46:08.000 Characteristics. Um, so they can meet the different neuromorphic compute… computing scenarios. Also, in this project, we hope to build the. 00:46:08.000 --> 00:46:19.000 Neuromorphic computing network system using such peptide-based memory strter. So, one of the important features we want to try is, in such a network. 00:46:19.000 --> 00:46:28.000 The interaction among the different synaptic nodes is driven by the ionic coupling, and instead of the traditional. 00:46:28.000 --> 00:46:35.000 Charger control processes. So this will be very different traditional… there's a similar space, uh. 00:46:35.000 --> 00:46:46.000 Memoristive integrated circuits. So, but to achieve such a goal. We need to address the important knowledge gap. The key question is. 00:46:46.000 --> 00:46:53.000 What kind of peptide nanostructures or assemblies that can be really used to make memory stirs? 00:46:53.000 --> 00:46:59.000 That can result in the low energy consumption for operation. Or. 00:46:59.000 --> 00:47:08.000 Uh, how do the different peptide, uh, structure. Influence the transport behaviors of the carriers and the. 00:47:08.000 --> 00:47:18.000 Ayers, and the result in the low energy consumption. So, to address such question, we have a bunch of the research activity proposed. 00:47:18.000 --> 00:47:28.000 I'll grant the next slide, please. So, first of all, we're gonna assemble the materials developed in the CRAS-1, the Thras 2. 00:47:28.000 --> 00:47:37.000 Into the memory stored device structure, and explore. How the different macromolecular assembly structure affect the. 00:47:37.000 --> 00:47:47.000 Ma'am resistive switching behaviors of such device. And we're gonna… using the voltage program and the pulse program method. 00:47:47.000 --> 00:47:54.000 To study the characteristics of such device and evaluate the device consumption. 00:47:54.000 --> 00:47:59.000 And also, we will… we will study the dynamic characteristics of the such devices. 00:47:59.000 --> 00:48:06.000 Your response to the spiking signal. So there are a bunch of the. 00:48:06.000 --> 00:48:12.000 Benchmark testing we're gonna perform, like a paired policy. Long-term potential agent, depression. 00:48:12.000 --> 00:48:17.000 And more specifically, in this project, we're gonna use a pep toy. 00:48:17.000 --> 00:48:26.000 Materials to build, uh, for example, the. The non-volatile memory stirs. So that's memory stirs with a long-term. 00:48:26.000 --> 00:48:32.000 Memory capability. So we're gonna test its performance at a synaptic. 00:48:32.000 --> 00:48:36.000 Components in the hardware-based. It's the biking neural network. 00:48:36.000 --> 00:48:41.000 To see how good this kind of device can handle the spatial, temporal data. 00:48:41.000 --> 00:48:51.000 And also, we're gonna make the memory stirs with, uh. Short-term memory, or the memory starts with very dynamic. 00:48:51.000 --> 00:49:05.000 And response characteristics, in response to, like, spiking signal. And, um, house-based, uh, real-time signal, and we'll be tested for processing the. 00:49:05.000 --> 00:49:15.000 The spatial-temporal information. Carried by the real-time analog signal. So this is important for the edge computing and robotic control application. 00:49:15.000 --> 00:49:25.000 Grand Next Place, next…. But yeah, thank you. And so we're going to use a method like nanofabrication, nanolithography, to make a nanosca. 00:49:25.000 --> 00:49:34.000 And the memory stirs involving the different peptide nanostructure, and we're going to perform the electronic ionic transport. 00:49:34.000 --> 00:49:38.000 Characterizations in combination with the materials characterizations. 00:49:38.000 --> 00:49:41.000 Why it is so small. 00:49:41.000 --> 00:49:48.000 Oh, we will also build the test stand, uh, the neural network systems. 00:49:48.000 --> 00:49:56.000 And he incorporated the Peptoi memory stirs. And either testing at a synaptic component, or. 00:49:56.000 --> 00:50:04.000 Testing as dynamic memory structure to study their response to the real-time signal or spanking-based signal. 00:50:04.000 --> 00:50:11.000 Uh, next page, please. So in my lab at the University of Michigan. 00:50:11.000 --> 00:50:17.000 And recently, uh, my PhD student established this, uh, called Memory Store-based. 00:50:17.000 --> 00:50:25.000 Rather computing networks. And right now, we are using the 2D layered materials, and in this project, we're gonna…. 00:50:25.000 --> 00:50:32.000 Incorporate the peptide-based memory registers. And to test the performance in the…. 00:50:32.000 --> 00:50:39.000 In the neuromorphic computing scenarios. So here, the red of our computing is a mathematical framework. 00:50:39.000 --> 00:50:47.000 For building the recurrent neural network. And used for parallel computing and control applications. 00:50:47.000 --> 00:50:57.000 So they take the dynamic signal, and without the… any, uh, the digital electronic process or involvement of the software. 00:50:57.000 --> 00:51:05.000 And such a network can directly generate to the anticipated control signals or some other output. 00:51:05.000 --> 00:51:15.000 For the, uh, parallel computing applications. So, traditionally, this kind of, uh, the red-work computing is implemented. 00:51:15.000 --> 00:51:27.000 In the form of the software, or the firmware. And… but here, we're using the memories to build all hardware-based radar computing network. 00:51:27.000 --> 00:51:33.000 So compared with a soft tool. Software-based radar computing network. 00:51:33.000 --> 00:51:42.000 Such a hardware devices can enable much lower energy consumption. And a much smaller footprint. So it's very good for the…. 00:51:42.000 --> 00:51:48.000 The system integration. And, uh, next… next slide… slides, please. 00:51:48.000 --> 00:51:56.000 And Grant, can you play the left video? So here, I'd like to show you, using our current radar computing. 00:51:56.000 --> 00:52:05.000 And we can demonstrate, uh, several, uh, different robotic control. Uh, control, uh, processes. 00:52:05.000 --> 00:52:10.000 For example, if the left video here, and yeah, if you click the…. 00:52:10.000 --> 00:52:19.000 I assume, yeah, here. So this is a target tracking. And navigation of a robotic rover. So during this process, you see the position. 00:52:19.000 --> 00:52:25.000 Signal of the target is directly sent to the right of our computing network. 00:52:25.000 --> 00:52:34.000 And almost no involvement of the software or digital electronics. So the rest of our computing layer directly. 00:52:34.000 --> 00:52:39.000 Generate the control signal. To control the movement of the rover. 00:52:39.000 --> 00:52:46.000 And Grandpa plays, uh. Later, yeah. Another one is my students want to use such. 00:52:46.000 --> 00:52:54.000 Rather computing, uh, layer to control a drone, but it's too challenging, so right now they decided to just control one arm of the drone. 00:52:54.000 --> 00:53:03.000 So, here you can see, um. Here to, like, a balancing bar using a propeller to control the position. 00:53:03.000 --> 00:53:13.000 So during this control process, the partition signal, the real-time signal, is direct, physically applied to the right of our computing process… the layer. 00:53:13.000 --> 00:53:19.000 And the regular computing layer, and you. Completely yielding the physical process. 00:53:19.000 --> 00:53:23.000 To generate the control signal, to keep the system in the balance. 00:53:23.000 --> 00:53:31.000 Uh, Grant, can you go back to the previous slide? I want to emphasize one thing. 00:53:31.000 --> 00:53:39.000 Yes, thank you. So… so in this project, we… we… the reason we choose a random computing network is. 00:53:39.000 --> 00:53:47.000 If you look at this schematic view in the blue color part, there's a reservoir layer, so it's a very dynamic process. 00:53:47.000 --> 00:53:55.000 To take the input signal. And so here, we need a short-term memory source with nonlinear switching behaviors. 00:53:55.000 --> 00:54:00.000 And the readout layer, the orange part, to generate the final output signal. 00:54:00.000 --> 00:54:09.000 So we need a long-term memory stirs with linear switching behavior. So it's a very good scenario. We can test the. 00:54:09.000 --> 00:54:16.000 Peptide-based memory registers. Because it's this kind of versatile chemistry of the peptide materials. 00:54:16.000 --> 00:54:21.000 Give us the opportunity, we can build the short-term and the long-term memory registers. 00:54:21.000 --> 00:54:27.000 To me, to the different, uh, the neuromorphic computing, uh, the scenarios. 00:54:27.000 --> 00:54:31.000 I saw there's a question from Ryan. 00:54:31.000 --> 00:54:36.000 Yes, thank you. So, my experience before with reservoir compute is for a control system and tokamax. 00:54:36.000 --> 00:54:46.000 And the key was that they used, like, um…. Training, like, a lot of training to try and get the weights right and the connections. 00:54:46.000 --> 00:54:52.000 And training ended up being, like, a genetic algorithm for the random connectivity. 00:54:52.000 --> 00:54:55.000 But is this continuously training itself? 00:54:55.000 --> 00:55:00.000 And for our, right, we're computing, we don't need to train the radar layer. 00:55:00.000 --> 00:55:06.000 So it's a completely rely on the… a little bit of randomness of the devices. 00:55:06.000 --> 00:55:07.000 And…. Yeah, it's random. 00:55:07.000 --> 00:55:11.000 Yeah. So, like, all the resistances start out random, and then they burn in over time, is that right? 00:55:11.000 --> 00:55:24.000 Yeah, they just needed a nonlinear response behavior that can. Can map the kinetic information carried by the input signal to a high-dimensional data space. That's good enough. 00:55:24.000 --> 00:55:25.000 Yep. Yep. 00:55:25.000 --> 00:55:35.000 So, if the reservoir state vector is diversified enough. So the readout layer can map to what we want for the anticipated control signal. So we only need to. 00:55:35.000 --> 00:55:36.000 I see. 00:55:36.000 --> 00:55:42.000 Train the readout layer, so that's a long-term memory source with a linear switching behaviors. 00:55:42.000 --> 00:55:49.000 Ah, cool. So the… did you say that the reservoir is the short-term memristers? 00:55:49.000 --> 00:55:50.000 Okay, good, cool. No, that's brilliant. It's an autonomous system. 00:55:50.000 --> 00:56:00.000 Exactly, yes. We need a more time. Yeah, thank you. So, we needed this very fast update rate, so we can use for the real-time application. 00:56:00.000 --> 00:56:03.000 Yes. Yeah, okay, I got it. That's wonderful, thank you. 00:56:03.000 --> 00:56:09.000 Thank you. Yeah. Grant, next slide, please. 00:56:09.000 --> 00:56:14.000 Yeah, next one. 00:56:14.000 --> 00:56:21.000 Yeah, right now, um, your… we work closely with PNNL. We are the garden, the Peptide. 00:56:21.000 --> 00:56:32.000 Nano, uh, nanostructure samples from PNRL collaborators. Like a peptide nanosheath, nanorod, nanotube, so we…. 00:56:32.000 --> 00:56:39.000 Right now, we have been making the memories using such material. So here, I show you the just example. 00:56:39.000 --> 00:56:47.000 Of the memory stores using the multi-nanotube channel, and also here you can see there's a memory steroid. 00:56:47.000 --> 00:56:52.000 We try to use a single nanotube. But right now, our fabrication process is quite a…. 00:56:52.000 --> 00:57:00.000 Uh, by chance, it's quite random, but it's good enough. For individual device fabrication right now, but later we're gonna propose. 00:57:00.000 --> 00:57:05.000 The research to develop methods, we can generate an array of such. 00:57:05.000 --> 00:57:16.000 Memory is jerse. Uh, next slide, please. And my student already performed some preliminary characterization, and especially. 00:57:16.000 --> 00:57:24.000 For the nanotube memory stirs. And we already see this hysteresis, IV characteristics that indicates. 00:57:24.000 --> 00:57:30.000 There's the memory still switching behaviors between the high resistance and low resistance states. 00:57:30.000 --> 00:57:37.000 So here, we said that there's a set compliant current limit, so you can see there's a rough change. 00:57:37.000 --> 00:57:44.000 Um, we haven't optimized this IV voltage, uh, this programming process. 00:57:44.000 --> 00:57:49.000 We hope we can further make a… make it a more reliable switching cycles. 00:57:49.000 --> 00:57:54.000 And we also performed this pulse program. Characterization, try to use the. 00:57:54.000 --> 00:58:00.000 The voltage pulses to tune the conductant states. We see the change of the conductant states, but. 00:58:00.000 --> 00:58:05.000 We haven't got a very neat. This asset recites courses. 00:58:05.000 --> 00:58:10.000 Um, so we're still working on that. Hopefully, next time, I can show you a much better. 00:58:10.000 --> 00:58:14.000 Let's go up, go down, this kind of pulse program courses. 00:58:14.000 --> 00:58:22.000 A nexus-wise place. Okay, there's also a result from the Lawrence Berkey Natural Lab. 00:58:22.000 --> 00:58:28.000 Um, I don't know if the Dr. Hsu or Dr. John is here like to present. 00:58:28.000 --> 00:58:34.000 No, I think both of them are not here. I mean, I can also briefly mention this one, it's fine, yeah. 00:58:34.000 --> 00:58:35.000 Thank you. Yeah, good luck. 00:58:35.000 --> 00:58:40.000 So here, basically. Uh, they're making, uh, this vertical device. 00:58:40.000 --> 00:58:48.000 So the result for this slide is to show. Even though, like, on the left, right, it's a silk, I mean, for this project, we don't really use silk. 00:58:48.000 --> 00:58:57.000 But this is a proof of concept to show. Because Silk has some ordering, and then when they compare to this amorphous peptide material. 00:58:57.000 --> 00:59:05.000 Even though both have memorized, uh. Behavior right there, ordering actually can make the performance much better, which is…. 00:59:05.000 --> 00:59:12.000 As we expected, because. Now we are going to use this capability to start to measure more this, uh. 00:59:12.000 --> 00:59:23.000 Self-assembled crystalline peptide material, including those. Peptide, like, the palm containing peptides I showed in the SRS2, yeah. 00:59:23.000 --> 00:59:24.000 I think the next slide. 00:59:24.000 --> 00:59:33.000 Yes, that's true. So, in the near future, and we're gonna do several things. First, we further improve the. 00:59:33.000 --> 00:59:43.000 Our programming and the setting, uh, the processes. To figure out the device physics for operating the peptide-based memory strips. 00:59:43.000 --> 00:59:52.000 And we will do more of the material characterization, and especially we want to visualize the kinetic… the behaviors of ions. 00:59:52.000 --> 00:59:57.000 And… or the formation of the ionic filaments in the peptide nanotube. 00:59:57.000 --> 01:00:02.000 And also, we want to… I've just mentioned, we want to develop the upscalable approaches. 01:00:02.000 --> 01:00:10.000 That can produce the larger arrays of peptide-based memories, not just random devices on the substrate. 01:00:10.000 --> 01:00:19.000 And also, like, the trend line and Grant mentioned. And we're gonna start the interface between the peptide nanostructure and the. 01:00:19.000 --> 01:00:28.000 You know, organic two-dimensional layer materials. Um, there could be a new opportunity to generate the memory stores with high uniformity. 01:00:28.000 --> 01:00:37.000 Next slide, please. So this is the approach I've mentioned we proposed this nano and macro printing process. 01:00:37.000 --> 01:00:46.000 We hope to produce the array. And of peptide-based memorister. So we start with, uh, PDMS template. 01:00:46.000 --> 01:00:52.000 The protrusive features define the device array. Then using the shear direction printing approach. 01:00:52.000 --> 01:01:00.000 We want to selectively deposit. Type to a nanotube on the protrusive surfaces, on the. 01:01:00.000 --> 01:01:08.000 The template. So after printing. We can stand out the arrays of such nanotube structure. Finally. 01:01:08.000 --> 01:01:14.000 After metalization, we can. Have the larger array of such devices. 01:01:14.000 --> 01:01:19.000 I'll grant the next page, please. And also, uh, we will…. 01:01:19.000 --> 01:01:28.000 I mentioned, in addition to the memory stirs. Uh, based on the pure peptide nanostructure, we also want to study the. 01:01:28.000 --> 01:01:36.000 The hybrid structure, right, integrate. And 2D layer the materials and the pipe toy assemblies. 01:01:36.000 --> 01:01:40.000 So here's one example. Right now, we already start to work on that. 01:01:40.000 --> 01:01:46.000 So, recently, my students developed this… this vertically arranged bismos cyanide. 01:01:46.000 --> 01:01:59.000 And also, we have a modern disulfide-based memories. And such a memory store has quite a unique switching behavior. Uh, grant the next page, please. 01:01:59.000 --> 01:02:07.000 So, here you can see, this is a set reset behavior of our bismos cyanide memorister, this vertically arranged device. 01:02:07.000 --> 01:02:12.000 And using the voltage pulses. We can realize anal. 01:02:12.000 --> 01:02:17.000 Conductance tuning. But more important, you can see here, once we stop. 01:02:17.000 --> 01:02:24.000 The application of the voltage pulses. Stop the sighting. The device is going to hold the conductance days. 01:02:24.000 --> 01:02:31.000 For a really long time. So that indicate really stable non-volatile retention of the conductance days. 01:02:31.000 --> 01:02:38.000 And also, you found this result, you can see there's almost no post-setting relaxation, because many. 01:02:38.000 --> 01:02:45.000 Reported analog memory stores after setting, there is a significant relaxation of the. 01:02:45.000 --> 01:02:51.000 The conductance. And so sometimes people need to integrate such memory store. 01:02:51.000 --> 01:02:58.000 With a current regulator. For example, our current limiting transistor to precisely set the. 01:02:58.000 --> 01:03:08.000 Conductance days we want to set a synaptic. Notes value. So, for such device, we have a… have a potential… we could build the…. 01:03:08.000 --> 01:03:14.000 The memory store network. Without such current regulator. 01:03:14.000 --> 01:03:21.000 And… but right now, the challenge is…. And such devices still exhibits significant device-to-device. 01:03:21.000 --> 01:03:27.000 The consistency, uh, issue. Next page, please. 01:03:27.000 --> 01:03:31.000 So, right now, we're doing this. So in my lab. 01:03:31.000 --> 01:03:41.000 We deposit the button electrodes. And deposited bitum and stylinite, we called it half-device sample, and we sent. 01:03:41.000 --> 01:03:47.000 Such a sample to PNL. Um, at the P&L, um…. 01:03:47.000 --> 01:04:00.000 The collaborator will deposit a PEP2 nanostructures, nanotube, nanorod. Also different poem structures, like Grant mentioned, soft landing, the iron soft landing. 01:04:00.000 --> 01:04:09.000 And Trend Long mentioned different morphology. And we want to incorporate the long-range order in such a structure. 01:04:09.000 --> 01:04:14.000 So we expect, and after fabrication, so when the sample go back to my lab. 01:04:14.000 --> 01:04:21.000 We're gonna finish the device structure. So, in search device, we expect the incorporation of the PEP2 nanostructure. 01:04:21.000 --> 01:04:27.000 Can regulate the formation site of the filament. So, we expect to see. 01:04:27.000 --> 01:04:34.000 Uh, you improve the device-to-device, uh, uniformity consistency, so that's good for our future. 01:04:34.000 --> 01:04:40.000 And the system, the integration. Yeah. So, uh, I will stop here, and…. 01:04:40.000 --> 01:04:53.000 Grant, how about… let's just end up here. Let's see if others have questions. I know, like, we kind of started late. 01:04:53.000 --> 01:05:03.000 Like, any… any question for us? 01:05:03.000 --> 01:05:04.000 Yes. 01:05:04.000 --> 01:05:12.000 I had a question, can you hear me? Hi, this is Paul. Um, sorry, I had some audio troubles, so I'm using my phone. 01:05:12.000 --> 01:05:17.000 Uh, hope you can hear the question. Basically, I was wondering, um. 01:05:17.000 --> 01:05:25.000 To what extent are you planning to, um. Look at the heterogeneity of properties of these. 01:05:25.000 --> 01:05:32.000 Structures. Um, you know, they… I'm reminded, for some of the, um, later slides on. 01:05:32.000 --> 01:05:40.000 The electrical characterization and trying to, uh. Develop more ordered, um, arrays of, um. 01:05:40.000 --> 01:05:45.000 Of wire-like, uh…. Uh… bomb structures. 01:05:45.000 --> 01:05:55.000 I'm reminded of carbon nanotube research, where. There's a pretty, uh, lengthy history of characterizing the variability. 01:05:55.000 --> 01:06:01.000 From, you know, properties from one wire to another, depending on its structure, or…. 01:06:01.000 --> 01:06:16.000 Uh, the way that it's bundled with other wires. Um, is there… is that kind of activity planned to sort of sort through these structures and see how variable their properties are? 01:06:16.000 --> 01:06:23.000 Yeah, once we get into this area we want to, uh, evaluate the variation from the device to device. 01:06:23.000 --> 01:06:29.000 Um, but my understanding, there's a peptide, like the nanotube, um, maybe it's different from the. 01:06:29.000 --> 01:06:36.000 I know the previous… there's research about assemble the…. Uh, like a cargo nanotube, but. 01:06:36.000 --> 01:06:43.000 Um, but I expect maybe the trend load can answer questions of how to control the, maybe the uniformity of the…. 01:06:43.000 --> 01:06:51.000 Yeah, the nice thing for this system is. Right? Like, when we started this peptide nanotube, or the nanosheet, we really started with, uh. 01:06:51.000 --> 01:07:00.000 This molecule self-assembly, right, we can…. Precisely control the structure, the crystalline material. 01:07:00.000 --> 01:07:06.000 And then we can keep, actually, the same, uh…. Morphology, right? But then tuning their chemistry. 01:07:06.000 --> 01:07:11.000 So now you can imagine, right, if we build a device with exactly the same. 01:07:11.000 --> 01:07:18.000 Kind of two, but different chemistry. I believe, you know, from that aspect, right, we can kind of, like. 01:07:18.000 --> 01:07:26.000 Adjust the challenge coming from the carbon nanotube, right? Because in the carbon nanotube synthesis, there might be some hedge…. 01:07:26.000 --> 01:07:31.000 Generally over there, but here, like, since we are doing molecular self-assembly approach. 01:07:31.000 --> 01:07:40.000 We know our, uh, crystal material system pretty well. And then once we know the performance from the device, we can actually. 01:07:40.000 --> 01:07:44.000 Narrowed down to specific tube, right, and then just keep the same…. 01:07:44.000 --> 01:07:52.000 Methodology and change the chemistry, see how the. Chemistry, as we kind of proposed, you know, to tune in the…. 01:07:52.000 --> 01:07:58.000 Like, the decoration of the ion transport, yeah. I don't know, Paul, did we answer your question? 01:07:58.000 --> 01:08:02.000 Okay, thank you. Yeah, thank you. 01:08:02.000 --> 01:08:08.000 Okay, thanks, yeah. I know some of you might need to go, right? But if you…. 01:08:08.000 --> 01:08:14.000 I'd like to ask other questions, you know, happy to…. Honestly, uh, yeah. 01:08:14.000 --> 01:08:20.000 Look, I have to drop off at 4, but you and Zhao Gunn are welcome to stay and take further questions. 01:08:20.000 --> 01:08:28.000 Okay, see if there are any more questions. And then we kind of start late, yeah, so…. 01:08:28.000 --> 01:08:30.000 I have to jump, but thank you very much. 01:08:30.000 --> 01:08:31.000 Okay, thank you. Yeah. I think if no more questions, then we would just end. 01:08:31.000 --> 01:08:37.000 Thank you. 01:08:37.000 --> 01:08:38.000 Yeah. Okay, thanks everyone. 01:08:38.000 --> 01:08:40.000 Yeah. 01:08:40.000 --> 01:08:46.000 See you