WEBVTT 1 00:04:51.330 --> 00:04:52.730 Paul McIntyre: Hey! There! Everybody! 2 00:04:54.050 --> 00:04:55.200 mark brongersma: At all. 3 00:04:55.200 --> 00:04:56.220 Wilman: Oh, hey! Paul! 4 00:05:09.540 --> 00:05:13.389 Paul McIntyre: Well, we'll wait for a couple of minutes before we get started. 5 00:05:52.610 --> 00:05:53.669 Jeff: Hey folks. 6 00:05:54.620 --> 00:05:55.490 Paul McIntyre: Hey, there. 7 00:06:18.460 --> 00:06:23.869 Paul McIntyre: wait another minute, just for for a few more people to join. 8 00:06:26.040 --> 00:06:28.889 Paul McIntyre: I'll go ahead and share my screen. In the meantime. 9 00:06:42.060 --> 00:06:43.200 mark brongersma: Looks, good. 10 00:06:43.670 --> 00:06:44.750 Paul McIntyre: Okay. 11 00:06:47.790 --> 00:06:49.110 Paul McIntyre: alright. 12 00:06:53.260 --> 00:07:17.989 Paul McIntyre: Well, we're stuck at 26 for a few seconds here. So maybe I should get started and we'll see if any others join. So we had last week, the 1st in a series of weekly presentations, summarizing the various teams that make up and projects that make up the Meerkat Microelectronics 13 00:07:18.100 --> 00:07:28.210 Paul McIntyre: Science Research Center. The 1st presentation in the series was given by Maurice the Lead Pi for the 14 00:07:28.210 --> 00:07:52.379 Paul McIntyre: the Nano Hybrids project led out of Lbnl. And so I'll be presenting this week. I'm Paul Mcintyre. I'm affiliated both with slack National Accelerator Lab and with Stanford University, and I'll be talking about a project called Esteem. 15 00:07:52.380 --> 00:08:17.300 Paul McIntyre: say a little bit more about the team, and all that in a moment. But the objective of this presentation, similar to Maurice's, I think, was mainly to introduce people across the various meerkat teams to the individual projects, some of the capabilities and ongoing efforts that we have and that we're building on 16 00:08:17.300 --> 00:08:36.269 Paul McIntyre: in our new research as part of this new project, and where we want to go with it in the future. And I'm in the perhaps unenviable position of having to represent research by colleagues from institutions that 17 00:08:36.270 --> 00:08:55.829 Paul McIntyre: various places around the country, and some of whom are on the line and can help me out if I make mistakes, but others are not here, so I apologize in advance, for if I misrepresent what they're doing, but I'll try to do my best to at least give a flavor of what we're working on. 18 00:08:56.620 --> 00:09:20.810 Paul McIntyre: Let me go ahead. So esteem stands for enabling science, for transformative energy, efficient microelectronics. It has a thrust and crosscut structure which is depicted here. In this figure there are 2 themes or thrusts that are centered on different hardware 19 00:09:20.810 --> 00:09:34.659 Paul McIntyre: centric approaches for reducing the energy footprint of microelectronics. One is materials and devices that merge memory and logic in 3D. 20 00:09:34.740 --> 00:09:56.550 Paul McIntyre: The other thrust is novel nanophotonic concepts for energy, efficient computing, sensing, and communication. And then there are 2 cross-cutting efforts, a co-design crosscut and Characterization crosscut. The institutions involved are slack. Stanford, Georgia, Tech Northwestern University. 21 00:09:56.550 --> 00:10:19.990 Paul McIntyre: the University of Tennessee and the University of Texas at San Antonio, and there's a lot of linkages among the crosscuts and the thrusts, as you would imagine. We have a pretty active schedule of meetings every 2 weeks for the 2 thrusts and sort of a monthly meeting at the crosscut level. 22 00:10:19.990 --> 00:10:43.040 Paul McIntyre: and the crosscut Pis are invited to attend the thrust meetings, and vice versa, to try to increase the amount of sort of cross fertilization of ideas, and we invite the students and postdocs and other personnel working in addition to the pis to attend all of these meetings. 23 00:10:43.040 --> 00:10:46.129 Paul McIntyre: we're trying to create a fairly. 24 00:10:46.390 --> 00:10:49.859 Paul McIntyre: a fairly well aligned, well integrated effort. 25 00:10:50.530 --> 00:11:15.320 Paul McIntyre: So let me say a little bit about these 2 thrusts. And, as I said, these are hardware approaches to enhance energy efficiency of microelectronics. And they basically focus on a fundamental problem of the way that microelectronics, microelectronic systems are structured today where there's a huge amount of energy wasted and a real performance limitation associated with 26 00:11:15.480 --> 00:11:31.849 Paul McIntyre: shuttling data back and forth between logic and memory, separate logic and memory chips. And this is a very wasteful process. So 2 approaches that we are focusing on in esteem. 27 00:11:31.850 --> 00:11:56.769 Paul McIntyre: and that each of these thrusts is devoted to are 3D. Monolithic integration, where we try to combine memory and logic functionality through vertical stacking, if you will, and chip to chip optical interconnection. So making interconnection from one chip to another, or from the chip chip to the cloud, using optical means 28 00:11:56.770 --> 00:12:02.200 Paul McIntyre: to try to surpass what electronic interconnection can achieve. 29 00:12:03.020 --> 00:12:31.559 Paul McIntyre: So the 1st approach, the approach of thrust one requires a number of things, including new memory and semiconductor materials synthesized at low thermal budgets. And so this actually turns out to be an interesting problem of how to discover and optimize materials to achieve high quality, memory and logic operations, even though they're synthesized at temperatures far lower than would be traditional in the semiconduct. 30 00:12:31.560 --> 00:12:32.850 Francois Leonard: I did not find him. 31 00:12:33.490 --> 00:12:39.149 Paul McIntyre: If you're not, speak, if you're not speaking, or if you're not presenting or or asking a question, please mute yourself. 32 00:12:40.290 --> 00:12:41.270 Paul McIntyre: Thank you. 33 00:12:42.390 --> 00:13:08.490 Paul McIntyre: The other thrust requires Cmos, compatible light sources and new approaches for light optical modulation integrated on ship. So both of them have this element of materials, discovery and optimization and device design that has a boundary condition 34 00:13:08.490 --> 00:13:13.370 Paul McIntyre: associated with it low thermal budgets or on-chip compatibility. 35 00:13:14.140 --> 00:13:39.660 Paul McIntyre: These 2 thrusts operate in tandem with a co-design approach that is informed by advanced characterization. So if you think about the more materials and device related elements of what we're doing in esteem, we can think about the integration of theory, modeling and simulation. So we have physics-based modeling informed by experiment. 36 00:13:39.660 --> 00:14:01.140 Paul McIntyre: including electronic and atomistic simulation tools, we need to do forward and inverse design for different kinds of bulk, thin film and heterostructure, architectures and materials, data-driven models that allow us to simulate system, behavior and optimization and uncertainty, quantification. 37 00:14:01.140 --> 00:14:24.479 Paul McIntyre: And then, in addition to that, we want to be able to ultimately get to sort of autonomous acceleration platforms. So the ability to do multi-scale and fidelity data fusion, active learning and guided data acquisition for discovery basically reduce the amount of time required for a useful scientific or engineering result. 38 00:14:24.630 --> 00:14:41.840 Paul McIntyre: and the combination of the co-design crosscut. And the characterization crosscut here is really essential. So I'm going to start off by introducing the co-design crosscut because it really plays a very central role in ultimately what we want to achieve with esteem. 39 00:14:41.840 --> 00:15:06.670 Paul McIntyre: There are 5 pis in this crosscut. Dirisha Kuditi. Pudi is one of those 5. She's an expert in neuromorphic computing, low power machine intelligence, brain inspired accelerators and neuro-inspired AI systems. So she's really the link between the sort of hardware software co-design and the ultimate application in terms of neuroscience inspired 40 00:15:07.750 --> 00:15:08.710 Paul McIntyre: applications. 41 00:15:09.040 --> 00:15:34.529 Paul McIntyre: Subashis. Mitra is from Stanford University is an expert in robust computing nanosystems, electronic system design and neurosciences. I should mention that Derisha is at Ut San Antonio. James Rondinelli is a professor at Northwestern University and an expert in electronic structure theory and materials. Co-design using 1st principles and machine learning methods. 42 00:15:34.530 --> 00:15:53.519 Paul McIntyre: Katie Schumann is at the Eecs department at the University of Tennessee. She's an expert in algorithm development applications, software and co-design approaches for neuromorphic systems. So she provides essentially a link to Derisha's activities, but also links back 43 00:15:53.520 --> 00:16:23.520 Paul McIntyre: to all of the activities of the other pis in the crosscut and then shiming. Yu is a faculty member at Georgia, Tech. He's an expert in crosslayer, co-design and combining device physics, circuit design and system benchmarking. One of the things that he is contributing specifically is to development of compact models for devices that then, can be can lead to more general results about circuit system level performance. 44 00:16:23.670 --> 00:16:49.199 Paul McIntyre: If we think about the crosscut topics, there's a complete co-design flow that we're trying to achieve in esteem through the activities of these 5 pis, basically at the materials level, predicting and designing properties of new electronic and photonic materials at the device level. Ml inferred device characteristics to optimize performance 45 00:16:49.200 --> 00:17:07.240 Paul McIntyre: in terms of integration. 3D integration of neuromorphic devices with data flow via photonics and then algorithms, lifelong learning for neuromorphic computing. That's getting to the ultimate sort of end application of where we hope to go with the team. 46 00:17:07.680 --> 00:17:36.410 Paul McIntyre: I'm going to give 2 quick exemplars of how the co-design crosscut is working and will work. Obviously I could have chosen quite a few different examples based on the work of the various pis. But in the interest of time I'm only going to show 2. 1 of them is courtesy of Shimeng Yu from Georgia, Tech, and I'll talk about the more of the materials and device relevant aspects of this 47 00:17:36.410 --> 00:18:00.309 Paul McIntyre: in a couple of slides. But this is a slide that basically talks through the simulation framework and methodology for understanding a particular new kind of memory that Shimeng is investigating along with others in esteem. And that's this ferroelectric, capacitive synapse charge domain in memory computing 48 00:18:00.410 --> 00:18:22.039 Paul McIntyre: you'll be familiar with our RAM arrays as an element for this kind of in-memory compute, and the reason to pursue this alternative approach. There's several different reasons. But one of the key ones is related to energy efficiency 49 00:18:22.040 --> 00:18:46.840 Paul McIntyre: in detecting the capacitance of a ferroelectric capacitive element in an array. We are really looking only at a dynamical event. So we're talking about dynamic power in understanding what the state of that memory element is in the case of a resistance change memory. There's both a dynamical power 50 00:18:46.840 --> 00:19:06.089 Paul McIntyre: component and a static power component involved in addressing the memory. And so that is one of the elements of this is a potentially more energy, efficient kind of compute memory compute. But in addition to that. 51 00:19:06.150 --> 00:19:13.179 Paul McIntyre: there's the ability to avoid one of the sort of difficulties of ferroelectric memory which is that it's a 52 00:19:13.220 --> 00:19:38.210 Paul McIntyre: it's a memory type that has to be is a destructive read, so it has to be rewritten in order to continue to read it, which is not the case here. So in any case, this kind of array of capacitive elements, each element will have a certain on-off ratio. There'll be some device to device variation in that. There'll also be noise. 53 00:19:38.210 --> 00:20:02.639 Paul McIntyre: both variational noise and thermal noise. And there's going to be charge transfer times that will differ from one element to another. So in the simulation workflow. The co-design workflow that Shimang has developed basically defines from data from individual capacitors. 54 00:20:02.640 --> 00:20:10.599 Paul McIntyre: What what a sort of a basis set of of input data is, then uses a 55 00:20:10.800 --> 00:20:35.749 Paul McIntyre: a computational method based on Gaussian sampling to basically create an optimized model for an array of such capacitors and extracts. The output signal latency, the variation in output noise, the effective number of bits, which is another way of saying the 56 00:20:35.750 --> 00:20:59.619 Paul McIntyre: accuracy with which this memory can operate, and then he repeats this process multiple times in order to achieve greater and greater fidelity of the model to the data. And then this is input into more of a circuit level simulation, looking at the entire array characteristic for things like image recognition. Using this 57 00:20:59.620 --> 00:21:23.389 Paul McIntyre: Vgg 8. Convolutional neural network for image recognition and a particular database called cfar 10 as an example. So basically testing, then, how effectively this capacitive memory array can perform relatively complex tasks in terms of image recognition and 58 00:21:23.390 --> 00:21:48.129 Paul McIntyre: similar kinds of applications. This is obviously going to be characterized in terms of accuracy, its dependence on device to device, variation on off ratio, and a lot of the energy related aspects of operating such an array. Future work will explore in-memory search with capacitive crossbars, and looking at 59 00:21:48.130 --> 00:22:11.610 Paul McIntyre: things like inverse matrix operations, to find Eigenvalues for Eigenvalue problems. Another example of what the co-design crosscut is doing and will do is in this lifelong learning topic that is being pushed forward by Professor Kudi Pudi. 60 00:22:11.610 --> 00:22:34.729 Paul McIntyre: and here she has developed an accelerator for continual learning at the edge, featuring systolic array architectures, and this is really part of an effort to understand bio-inspired, lifelong learning mechanisms, and how they can be performed in computation. 61 00:22:34.980 --> 00:22:59.899 Paul McIntyre: And an example of that is a synaptic neural network with task agnostic local plasticity that her group has been studying and involves a fairly complex co-design cycle and coupling to this lifelong learning accelerator that her group has been working on. So this is a capability that we hope to be able to leverage going forward 62 00:22:59.900 --> 00:23:24.869 Paul McIntyre: esteem. As we develop more of these new computing memory and neuromorphic device approaches and the ability to then understand their system, level performance through the work of the co-design crosscut returning to thrust one. So what are the new materials and devices that we want to study to merge 63 00:23:24.870 --> 00:23:48.280 Paul McIntyre: and memory in 3D. I've already talked about one of them capacitive synapses. And basically, this is a cartoon of what one of these capacitors looks like in cross-section, where there's an asymmetry in the electrodes on either side of the ferroelectric material. The ferroelectric material here would be hafnium, zirconium oxide. 64 00:23:48.280 --> 00:24:13.139 Paul McIntyre: And it's this asymmetry that gives rise to a distinguishable capacitance state, depending on how the ferroelectric material is polarized and that's what really sets the state of each memory cell in the Capacitor array. In addition to that, we're interested in electrochemical synapses. I'll say more about that in a moment spiking neurons 65 00:24:13.140 --> 00:24:38.099 Paul McIntyre: based on ferroelectric fets, oscillatory neurons. I'm not going to have time to talk about this, but it's another approach involving complex oxides that exhibit electrically triggered insulator metal phase transitions, and then the general topic of being able to make back end of line compatible logic. Ultimately Cmos, combining the already demonstrated 66 00:24:38.100 --> 00:24:59.749 Paul McIntyre: back into line compatible nmos devices which can be made based on atomic layer, deposited semiconductors, oxide semiconductors at quite low temperatures and ultimately developing candidate Pmos oxide semiconductors that are compatible and can be put together in a Cmos architecture. 67 00:25:00.100 --> 00:25:25.419 Paul McIntyre: So once again, this is the Shimeng's Capacitor array crossbar memories. The idea here is that we charge each capacitive synapse and then transfer the charges to a reference capacitor. And that's how the memory works and the difference in state of the memory is dependent, as I said, on the effectively the screening conditions provided by the 68 00:25:25.560 --> 00:25:50.480 Paul McIntyre: top and bottom electrodes. These are often different materials, perhaps with an inserted layer of a semiconductor like atomically deposited indium, tungsten oxide, and also may have a difference in the area of the electrodes, and this gives rise to discernible capacitance difference at 0 applied voltage which can be read to determine the state of the memory. So there's further research needed 69 00:25:50.480 --> 00:26:14.660 Paul McIntyre: to optimize the sort of the interface, physics and chemistry. In these memories, in addition to the more circuit and system level considerations I've already discussed. In addition to that electrochemical field effect. Transistors are of interest. These are really remarkable devices in the sense that they exhibit this. 70 00:26:14.660 --> 00:26:39.050 Paul McIntyre: This really significant multi-state programmability. They work basically very much like batteries do, in the sense that you're looking at insertion of ions into a material. In this case, insertion of ions into an organic semiconductor that modifies the conductance of that semiconductor channel. And this can be done with 71 00:26:39.050 --> 00:27:03.199 Paul McIntyre: really remarkably small amounts of energy input to shift the threshold voltage of the field effect transistor. And this is shown here in terms of the programmability for different molecular structures that affect the ease of proton exchange in and out 72 00:27:03.200 --> 00:27:28.140 Paul McIntyre: of the Channel. It's possible to develop double gate architectures, and these have been used by Professor Saleo's group. This is Alberto Saleo's work to demonstrate organic artificial neurons with tunable spike frequency adaptation. So there's a lot of interesting brain inspired computation that these materials and devices can 73 00:27:28.140 --> 00:27:29.820 Paul McIntyre: support. 74 00:27:30.030 --> 00:27:43.539 Paul McIntyre: Another effort in thrust, one also, looking at brain inspired computing. These are ferroelectric FET based spiking neural networks. And this is work led by Shumandata's group at Georgia, Tech. 75 00:27:43.760 --> 00:28:08.219 Paul McIntyre: And the idea here is to combine a ferroelectric FET, which has both a logic and a memory function with more standard silicon based devices in a synaptic neural network compute block including neurons and synapses. And this is then used to study low power spike based information process 76 00:28:08.220 --> 00:28:18.870 Paul McIntyre: processing which exploits this co-localized memory functionality in the computing in this kind of an array. 77 00:28:19.240 --> 00:28:43.730 Paul McIntyre: So this is another direction of taking advantage of the capabilities of the team in terms of ferroelectric thin film materials and finally work on indium based n-type semiconductors for back end of line transistors what we have found. And this is sort of work in parallel between my group and Professor Dada's group is that 78 00:28:43.730 --> 00:29:08.660 Paul McIntyre: the introduction of tungsten as a dopant into indium oxide can produce a remarkable change in indium oxide's semiconducting properties, and this is largely attributable to a change in the oxygen stoichiometry of the material, adding just a few atomic percent. Tungsten can greatly increase or greatly improve the oxygen stoichiometry and bring 79 00:29:08.660 --> 00:29:32.410 Paul McIntyre: it very close to the ideal in 2 0. 3, that one would expect, but which usually is not observed for indium oxide, thin films, and this change in oxygen. Stoichiometry corresponds to a much improved semiconductor performance, particularly the ability to shift the threshold voltage of these devices by many volts, and also achieve very steep slope 80 00:29:32.460 --> 00:29:47.680 Paul McIntyre: turn on behavior and while retaining quite high electron mobilities in the range of 20 to 50 typically, and what we found is that there's a strong correlation, not only of the tungsten 81 00:29:47.680 --> 00:30:11.679 Paul McIntyre: concentration with oxygen, stoichiometry, and with threshold voltage, but also with the degree to which there may be some longer range crystalline order. In this material these are usually thought of. Indium oxide is usually thought of as an amorphous oxide semiconductor. When it gets down to a sufficiently small film thickness of a few nanometers. But what we've actually found 82 00:30:11.680 --> 00:30:36.630 Paul McIntyre: using X-ray pair distribution function analysis is even very thin. Indium oxide films, a few nanometers thick, actually exhibit crystallinity, but the addition of just a few percent of tungsten suppresses that very strongly. We have more of an amorphous type structure. So the tungsten looks like a very interesting approach for modifying many of the structural electronic 83 00:30:36.630 --> 00:30:43.359 Paul McIntyre: of this material. And we're still trying to understand all of the details and how to optimize them. 84 00:30:44.396 --> 00:30:44.933 Paul McIntyre: So 85 00:30:45.520 --> 00:31:08.659 Paul McIntyre: moving on to thrust 2 chip to chip optical interconnects and sort of the materials that can help achieve those one of the main approaches that we're pursuing here in terms of new materials, is looking at Group 4 on-ship light emitters. So it's been known for quite some time that addition of small amounts of tin to Germanium 86 00:31:08.660 --> 00:31:32.049 Paul McIntyre: can alter the band structure of the material from an indirect gap. Semiconductor to a direct gap. Material and addition of silicon allows further tuning of the band gap. This is a challenging system, though, because the concentrations of tin required to achieve this advantageous band structure change 87 00:31:32.050 --> 00:31:55.529 Paul McIntyre: are far beyond the bulk solubility of tin in diamond, cubic Germanium. So there's a lot of crystal growth research that has happened and is needed to be able to actually achieve this direct gap behavior in this alloy system. So what we have done, having 88 00:31:55.530 --> 00:32:17.110 Paul McIntyre: spent some years working on crystal growth tricks that allow this to happen. Recently, we've been looking at the combined effects of the alloy composition which we can now tune over a wide range and built-in tensile strain in the mismatched epitaxial structures of silicon, Germanium, tin, and germanium. 89 00:32:17.110 --> 00:32:41.740 Paul McIntyre: an example of that is making freestanding nanowires. These are etched and released from an underlying layered substrate, and and doing that in such a way that we can actually create substantial tensile strain in the Germanium wire that we have produced, and then coating it with Germanium, tin or silicon Germanium tin. 90 00:32:41.740 --> 00:33:06.200 Paul McIntyre: which further enhances the strain in the material. The tensile strain and shifts the band gap further and makes it more direct. So this is an interesting platform for studying the combined effects of strain and composition in this highly mismatched system, and a system that's really a metastable composition and the ability to control these things 91 00:33:06.200 --> 00:33:08.290 Paul McIntyre: in an on-ship structure. 92 00:33:09.720 --> 00:33:31.880 Paul McIntyre: We're combining that with work with Yelena Vukovich's group and her group is working on inverse design of structures that will allow the coupling of these nanowires. These strained and composition controlled nanowires with the surrounding 93 00:33:31.880 --> 00:33:56.840 Paul McIntyre: chip structure in order to make an optimal Germanium tin cavity that will simultaneously support high performance, photonic and mechanical properties. That is the strain required to achieve a good direct gap and to tune the optical properties. So mechanical simulations that we've performed on these structures so far 94 00:33:56.840 --> 00:34:21.250 Paul McIntyre: suggest that the strain in the Germanium wire before we've coated it with the Germanium tin should be about 2% and tensile along the wire axis. That's a very large strain that is very helpful in terms of achieving that direct gap. So, combining these mechanical simulations with inverse simulations of light, coupling into and out of 95 00:34:21.250 --> 00:34:46.239 Paul McIntyre: the wire structures by making structures that look like this that have this sort of designed corrugation of the refractive index through patterning and etching allows for near 0 transmission. Therefore, maintaining the optical fields within the wire, and significant 96 00:34:46.239 --> 00:35:02.829 Paul McIntyre: reduced footprint compared to conventional designs. So this is a exciting combination of materials, growth and structure formation with inverse design of the optical elements required to create a good 97 00:35:02.830 --> 00:35:27.759 Paul McIntyre: light source. Ultimately, in addition to the light sources, we're interested in modulation, and how new materials can be helpful in achieving better on ship light modulation in a smaller footprint, so that one approach is to use materials that exhibit very large optical nonlinearities, barium, titan, 8 being a good example of that, it has about 50 times the poculs coefficient 98 00:35:27.760 --> 00:35:52.720 Paul McIntyre: of the standard lithium niobate electro optic modulators. The challenge has been making a high quality barium, tight knight material that can be integrated in a reasonable way in order to confine light and to do the other things that a modulator needs to do while coping with things like epitaxial mismatch and 99 00:35:52.860 --> 00:35:57.720 Paul McIntyre: the potential for forming dislocations and other defects during crystal growth. 100 00:35:57.850 --> 00:36:23.630 Paul McIntyre: So in this particular topic, the research group of Harold Wong at Stanford, and also Yelena Vukovich's group, are looking at barium titanate, thin films deposited on a particular orientation of a strontium, titanate substrate, that produce quasi-single domain 101 00:36:23.800 --> 00:36:46.630 Paul McIntyre: Bto thin films, and that enables maximal electro-optic response and precise polling to achieve quasi-phase matching and then using an amorphous silicon layer to create a wave guide that can allow these optical modes to propagate and to maintain an effective pockles coefficient for this kind of structure. 102 00:36:47.880 --> 00:37:12.860 Paul McIntyre: And finally, another aspect of thrust 2 is basically combining interesting surface structures, metasurface structures with different wave guides in order to basically steer light or program optical beam control in new ways. So this is work that's being pioneered by Mark Brongersma's group at 103 00:37:12.860 --> 00:37:37.100 Paul McIntyre: Stanford. A couple of different approaches are shown here schematically, and there are publications here describing results from these approaches. One is basically using integrated photonics, often fabricated of silicon nitride on a dielectric wave guide and being able to then couple light out of 104 00:37:37.100 --> 00:37:55.570 Paul McIntyre: grating couplers from the integrated photonics into patterned metasurface structures of various kinds, and thus manipulating the geometry of this interaction in such a way to have programmable output with different kinds of 105 00:37:55.570 --> 00:38:20.320 Paul McIntyre: beam patterns coming off of this surface. Another approach which involves plasmonics would use instead of a dielectric wave guide something like gold, which supports strong plasmonic resonances on its surface to couple optical fields from a source to a 106 00:38:20.320 --> 00:38:44.819 Paul McIntyre: patterned metasurface. And here there's a dielectric medium separating the absorber here, and the patterned structures on the surface that provides the required out coupling of the light and control. We could imagine replacing this aluminum oxide 107 00:38:44.840 --> 00:39:05.460 Paul McIntyre: ultra, thin dielectric, which is typically deposited by Ald with something like hafnium zirconium oxide, a fair electric which would achieve an additional degree of control by electrically programming the polarization and the nonlinear optical characteristics of the material dynamically, that's a future direction. We want to go in. 108 00:39:05.640 --> 00:39:29.839 Paul McIntyre: and I'll end here with just a few slides describing the characterization capabilities that underlie a lot of the materials, development and materials research that we're doing here as part of esteem. We have the good fortune of having 2 major X-ray light sources at slack. One of them, Ssrl is being used for a variety of different 109 00:39:29.980 --> 00:39:54.650 Paul McIntyre: tasks within this project, but one of them that's highlighted here is using a new beam line and some new instrumentation that we've developed around it to do either static or time. Resolved grazing incidents, synchrotron X-ray diffraction. And this is really to look at in situ experiments 110 00:39:54.650 --> 00:40:19.010 Paul McIntyre: with the ability to control thermal excursions, annealing treatments, etc, down to the sub millisecond time scale and to measure changes in the structure of materials as they evolve thermally at those time scales. An example shown here from a typical ferroelectric capacitor structure includes the 111 00:40:19.010 --> 00:40:43.830 Paul McIntyre: various crystalline layers that are present, and the interest here is being able to understand the formation of the ferroelectric phase of Hco, which occurs usually during an annealing process and the dynamics of that, and how it's related to the thermal treatment that's used. And with this flash lamp annealing system that has been developed by John Benecki, one of the pis who's actually leading this 112 00:40:43.840 --> 00:40:54.940 Paul McIntyre: characterization crosscut. It's possible to look at the time evolution of X-ray 113 00:40:54.940 --> 00:41:19.149 Paul McIntyre: signals, the diffractive peak intensity and how that's related to the pulse of the flash lamp annealing system and understand the dynamics of the crystallization event that produces the orthorhombic phase for electric material. It's also important to note here that we can use the coefficient of thermal expansion of these materials 114 00:41:19.150 --> 00:41:45.229 Paul McIntyre: to as effectively a thermometer to look at the effect of the thermal pulse and the time of the thermal pulse relative to the transformation time, and to do that with real accuracy in terms of the surface temperature of the sample. So this is really kind of a unique capability of being able to do thermometry, as we study phase, transformation kinetics using X-rays. 115 00:41:45.510 --> 00:41:53.880 Paul McIntyre: Colin Ofus is a faculty member at Stanford, and one of the real pioneers in the technique of four-dimensional 116 00:41:53.890 --> 00:42:18.850 Paul McIntyre: stem 4 D stem basically collecting a series of electron diffraction patterns using scanned electron probes. Each pattern has got a two-dimensional data set associated with it. And of course, it's scanned over a two-dimensional area. So hence we have 4 d. Stem, and by calculating the expected diffraction patterns for experiments 117 00:42:18.850 --> 00:42:42.470 Paul McIntyre: like this and comparing them to the actual measured patterns, it's possible to retrieve a lot of information about the sample, including phase information related to the passage of the electron beam through the sample, but also classification of the sample into different crystal structures and orientations, and also quantification of local strain states. 118 00:42:42.470 --> 00:43:07.450 Paul McIntyre: So the combination of the computational methods that Colin has developed, and some of the new tem instrumentation that we have allows this 4 D stem technique to be very powerful. An example of that is an example of one. Such technique is a nanobeam electron diffraction analysis of a thin ferroelectric film sandwiched between 2 thinetride electrodes. 119 00:43:07.450 --> 00:43:32.029 Paul McIntyre: And basically this is lighting up the different phases of the Hco, there are 4 different phases that are present in this sample, and you can see through the false color image there sort of volume fractions, but it also gives information about the orientation of each of these crystalline domains as well. So it's very powerful in terms of what it can tell us about these kinds of samples. 120 00:43:32.170 --> 00:43:56.199 Paul McIntyre: Finally, the ability to do tychographic phase, retrieval with methods like this allows much higher atomic resolution, much cleaner images than in a conventional stem image, where we can actually see in these images individual point defects or small clusters which are 121 00:43:56.200 --> 00:44:07.009 Paul McIntyre: very difficult to resolve in a normal image. So these kinds of techniques are ones that Collins group is really pioneering, but they're also ones in which he's working with various pis across the team 122 00:44:07.010 --> 00:44:31.999 Paul McIntyre: and finally combining some of the elements of what we've just seen, the ability to do, sort of high spatial resolution characterization with the in situ and operando opportunities that X-ray diffraction presents. We have at slack the Linac coherent light source. It's the world's 1st X-ray free electron laser, and this through the coherence of the 123 00:44:32.000 --> 00:44:55.810 Paul McIntyre: X-ray beam, allows for new kinds of X-ray imaging, a coherent X-ray diffraction imaging, but also in a format where one can really think about doing some sophisticated in situ and operando, 3. D imaging. So Ariana Gleeson at slack pi in this crosscut, is leading this effort at slack 124 00:44:55.810 --> 00:45:13.560 Paul McIntyre: and basically combining these elements of high coherence of the X-rays and the relatively non-destructive nature, and the ability to handle different sample environments, to lead in situ and operando. 3D. Imaging. 125 00:45:13.740 --> 00:45:29.709 Paul McIntyre: So that is a quick tour through esteem, and I apologize to any of my colleagues for misrepresenting their work. But I hope they'll correct me, and in the meantime I would welcome any questions. Thank you. 126 00:45:37.830 --> 00:45:44.370 Jeff: I was really good, Paul. Thanks for setting the standard. I got to up my game here on June 24.th 127 00:45:45.545 --> 00:45:51.400 Jeff: A 1 question. There's a lot of there's some overlap, of course, of materials and things 128 00:45:51.550 --> 00:45:57.470 Jeff: with our project. But the other thing I noticed especially, say, for, like the silicon Germanium tin. 129 00:45:57.650 --> 00:46:01.629 Jeff: There's awful lot of work that's going on here. There's an Efrc 130 00:46:01.790 --> 00:46:06.439 Jeff: that's ongoing for a couple years now around silicon Germanium 10. 131 00:46:06.590 --> 00:46:11.770 Jeff: And I wonder if there's some useful interactions there, if if that's. 132 00:46:11.770 --> 00:46:12.150 Paul McIntyre: Yeah. 133 00:46:12.150 --> 00:46:12.770 Jeff: Interested. 134 00:46:12.770 --> 00:46:13.660 Paul McIntyre: For sure, actually. 135 00:46:13.660 --> 00:46:17.849 Jeff: And a lot of growth and modeling and characterization going on. 136 00:46:17.850 --> 00:46:23.246 Paul McIntyre: Yeah, that's that's a very good question and an exciting project. I'm actually, I'm actually a pi in that. 137 00:46:23.530 --> 00:46:24.030 Jeff: Oh, sorry! 138 00:46:24.030 --> 00:46:47.680 Paul McIntyre: The Frc. So what we've my group's been doing is studying short range order, which is a interesting kind of additional wrinkle that occurs in alloy systems. So beyond the average composition and the strain, if you have short range order, there are modifications to the band structure that could be quite important. And 139 00:46:47.680 --> 00:47:12.410 Paul McIntyre: that center is really trying to understand those kinds of effects. So we've been using X-rays at Ssrl to measure short range order in samples. Not only that we deposit by Cbd. At Stanford, but also the other team members deposit through a variety of different methods. And so I think that we can leverage those 140 00:47:12.410 --> 00:47:16.410 Paul McIntyre: collaborations, and and, you know, look forward to doing that. 141 00:47:16.800 --> 00:47:19.849 Jeff: Okay, yeah, I'd forgotten you're all part of that. Sorry about that. 142 00:47:20.090 --> 00:47:20.850 Paul McIntyre: It's okay. 143 00:47:21.100 --> 00:47:21.860 Jeff: Yeah, yeah. 144 00:47:23.220 --> 00:47:30.149 James Ang: I'll add a couple of questions in terms of trying to find linkages. I was wondering if the 145 00:47:30.510 --> 00:47:38.569 James Ang: if the modeling and simulation capabilities that Jackie Yao presented last week 146 00:47:38.830 --> 00:47:43.639 James Ang: are are tools that your team would also be interested in using. 147 00:47:43.830 --> 00:47:50.000 James Ang: I don't have a vested interest in that, but it seemed like that might be a natural integration point. 148 00:47:51.030 --> 00:48:19.259 Paul McIntyre: Yeah, I think that's true. And I'm I'm not, I think, because of travel. Some of our co-design team members were unable to attend that presentation. But of course we have the recording, and we'll share that around with with all the members of the team, particularly Katie. Katie Schumann is on the line now, and I will encourage her to look through that with her colleagues on the, on the co-design 149 00:48:19.260 --> 00:48:25.960 Paul McIntyre: crosscut, and see where the opportunities are. I'm sure there are opportunities for collaboration there. 150 00:48:26.610 --> 00:48:41.090 James Ang: Terrific. And one of my team members or copis is Diana rosing from San Diego. She she's done a lot of work, a lot of background and history and lifelong learning. 151 00:48:41.450 --> 00:48:42.160 Paul McIntyre: Okay. 152 00:48:42.552 --> 00:48:55.517 James Ang: One of the capabilities he's bringing to our team is a hyper dimensional compute. And in particular, while there could be digital implementations we're interested in in some of the energy efficient 153 00:48:56.040 --> 00:49:01.760 James Ang: benefit energy efficiency benefits from trying to develop analog implementations. 154 00:49:01.760 --> 00:49:02.270 Paul McIntyre: Hmm. 155 00:49:02.270 --> 00:49:12.990 James Ang: And hopefully, that's another area of potential synergy and alignment between. your project and your team and and some of the work and and my team. 156 00:49:13.290 --> 00:49:36.989 Paul McIntyre: Yeah, for sure. I I think that I think we there's some things we could talk about there, both from on the cross design across the co-design Cross cut and also thrust one some of the devices that there have a really interesting analog programming characteristics. So we could. We should. We should look for opportunities to bring people together on that. 157 00:49:38.540 --> 00:49:39.987 James Ang: Terrific. And then 158 00:49:40.880 --> 00:49:51.870 James Ang: yeah, there, there may be some connections. I'll let Grezagorsk point out anything around your thrust. 2 and silicon optical networking and 159 00:49:52.030 --> 00:50:07.269 James Ang: integration with with logic. I was curious, though, in general, for thrust one and 2. Are you thinking about hybrid bonding as a packaging technology that you'll be trying to leverage and and make use of. 160 00:50:08.090 --> 00:50:30.910 Paul McIntyre: Yeah. So we're we're looking more at sort of for the for thrust one. This is really aiming at. Kind of monolithic. 3D integration. So not stacking so much as basically, you know, sort of growth and planarization related processes. 161 00:50:30.910 --> 00:50:32.619 James Ang: Back, but I wasn't sure if that was. 162 00:50:32.620 --> 00:50:54.830 Paul McIntyre: Yeah. But I think on the in terms of the photonic integration, there's a lot of potential there for I mean, in both cases, there's some things we probably would want to do truly, on ship, but other things where we might have bonding on interposers of various kinds. So I think there's room for different 163 00:50:54.830 --> 00:51:18.219 Paul McIntyre: design options. This is one of the topics where Subashi Mitra is quite valuable, because he's kind of got a foot both in the photonics camp and in the electronics camp and has done some interesting sort of co-design activities sort of looking at the trade-offs between different architectural choices. So yeah, that's 164 00:51:18.220 --> 00:51:23.539 Paul McIntyre: that's something that he would would have something more intelligent to say about than I would. 165 00:51:23.540 --> 00:51:27.409 James Ang: Okay? Well, you know, I'm I'm trying to look also for connections, not just 166 00:51:27.650 --> 00:51:33.620 James Ang: between projects within meerkat, but also with capabilities that the the 167 00:51:33.860 --> 00:51:38.089 James Ang: Chips and science program is establishing for their 168 00:51:38.270 --> 00:51:43.059 James Ang: their facilities, and in particular prototyping and advanced packaging facility. 169 00:51:43.430 --> 00:51:45.489 Paul McIntyre: Yeah, that's a good point. Yeah. 170 00:51:47.850 --> 00:51:48.810 James Ang: Great. Thank you. 171 00:51:49.040 --> 00:51:50.670 Paul McIntyre: Sure. Thank you. Thanks, Jim. 172 00:51:50.860 --> 00:51:53.570 Paul McIntyre: Any other, any other questions or comments. 173 00:52:04.670 --> 00:52:09.120 Paul McIntyre: Well, if not, do you feel we we will. 174 00:52:09.120 --> 00:52:20.759 Paul McIntyre: We are recording this, and we will make it available to the to the pis of each of the 8 teams within meerkat, and they will be 175 00:52:20.760 --> 00:52:39.550 Paul McIntyre: then, you know, making it available to the team, their team members. So you should. You have an opportunity to review it. And of course, if you have any, follow up questions, feel free to reach out, I'd be happy to try to answer them, or or connect you to people on our team who would be better positioned to answer them than I am. 176 00:52:39.600 --> 00:53:04.509 Paul McIntyre: We're assembling a schedule for the remaining 6 talks in this series. So look for an announcement soon about the next presentations that we have queued up and hope to see turnout like this at, or better, even at many of the future presentations. 177 00:53:05.810 --> 00:53:08.789 Jeff: Hey, Paul, I had a follow up question. Not on. 178 00:53:08.790 --> 00:53:09.389 Jeff: There's a team. 179 00:53:09.390 --> 00:53:11.839 Jeff: But I had sent everybody an email 180 00:53:12.040 --> 00:53:26.210 Jeff: in regards to the sent user meeting. If there was any interest, you know, in in 2 modes, one just coming and talking about your project or having somebody from your project. Come there. We're going to have 181 00:53:26.470 --> 00:53:30.270 Jeff: you know. Jennifer Hollingsworth is going to have the Chime center. 182 00:53:30.440 --> 00:53:30.910 Paul McIntyre: Hmm. 183 00:53:31.680 --> 00:53:38.450 Jeff: be part of the sent user meeting part of a symposium, and they're going to use it as part of their 184 00:53:38.590 --> 00:53:40.680 Jeff: annual meeting as well. 185 00:53:41.050 --> 00:54:01.659 Jeff: And I was, you know, I sent that email to everybody to see if there was one interest in doing something like that. And if not, is there interest in general, and just somebody, either yourself, folks, or somebody from your team coming talking about the project within meerkat? That was kind of what I was trying to. I'm trying to start to firm up the 186 00:54:01.830 --> 00:54:04.550 Jeff: set user meeting, which is in September with 187 00:54:05.100 --> 00:54:17.049 Jeff: speakers around Msrc. Since Jennifer's. They're gonna do all of theirs. My team will definitely come, but wondering if our folks want to participate in one way or the other. 188 00:54:18.815 --> 00:54:33.010 Paul McIntyre: My own. Take on it. I think it's a good idea, and I think the reason I didn't respond is that my own calendar is so crazy in September I'm still trying to figure it out. But I think in principle, yeah, it sounds like a good idea. 189 00:54:34.110 --> 00:54:36.080 Jeff: How about how about the rest? 190 00:54:36.740 --> 00:54:38.879 Jeff: Anybody else that's on here. 191 00:54:39.800 --> 00:54:43.372 Maurice Garcia-Sciveres: Yeah, I think. Oh, I think we have. Yeah, I have to 192 00:54:44.390 --> 00:54:48.420 Maurice Garcia-Sciveres: check. Who can make it? Also, it's an issue of dates as well. Yeah. 193 00:54:51.130 --> 00:54:51.600 James Ang: Yeah. 194 00:54:51.600 --> 00:54:51.930 Jeff: Online. 195 00:54:51.930 --> 00:55:02.550 James Ang: For me, Jeff. You know. I think I've told you before. My project is is not directly into in the materials and devices. Area. 196 00:55:02.550 --> 00:55:03.300 Jeff: Yeah, yeah. 197 00:55:03.300 --> 00:55:13.849 James Ang: As we pointed out, as I pointed out, just just you know, 5 min ago. There we do. We do have a couple of copis that might might be relevant and be interested. So 198 00:55:14.470 --> 00:55:15.080 James Ang: I'll have to. 199 00:55:15.080 --> 00:55:20.299 Jeff: Yeah. So that's kind of what I need, you know, not not too long from now is some sort of 200 00:55:20.680 --> 00:55:21.900 Jeff: commitment 201 00:55:22.720 --> 00:55:34.829 Jeff: from, you know, either team or or somebody from your team that would come and talk about it, because I gotta eventually put this schedule together for the user meeting and and see how that's all gonna work out. So 202 00:55:35.571 --> 00:55:46.830 Jeff: I I realize everybody's got pack schedule. But it doesn't have to be you necessarily unless we're using it as an annual meeting. I know we'll have to have an annual meeting at some point. 203 00:55:50.390 --> 00:55:53.410 Jeff: So appreciate, you know. Give some thought, and let me know 204 00:55:53.800 --> 00:55:57.969 Jeff: reasonably soon, so that I could start to firm up the firm up the schedule. 205 00:56:00.490 --> 00:56:11.809 Paul McIntyre: Okay, I'll I'll I'll do a huddle with our pis, and and we'll we'll get back to you on on that. Jeff, I I as I said, I think it's a good idea that somebody should go. 206 00:56:12.100 --> 00:56:22.319 Jeff: Okay, that'd be great. And we'll welcome students and all that, because we have, we have poster, big poster session that happens as part of the meeting as well. So there's lots of ways to do it. 207 00:56:23.140 --> 00:56:34.060 Paul McIntyre: Yeah, I think it's a good idea for our our students and postdocs to get a better idea, a better understanding of the capabilities at the now centers, including sin. 208 00:56:34.670 --> 00:56:36.790 Jeff: Yeah. And yeah, that'll be good. 209 00:56:37.020 --> 00:56:37.610 Paul McIntyre: Okay. 210 00:56:37.630 --> 00:56:39.029 Jeff: All right. Thanks, everybody. 211 00:56:41.131 --> 00:56:52.179 James Ang: By the way, you know, given the travel restrictions on all of our program managers will. Will that your meeting Jeff? Have a virtual online attendance, option. 212 00:56:52.820 --> 00:56:53.690 Jeff: No. 213 00:56:53.910 --> 00:56:55.080 James Ang: Yeah. Okay. 214 00:56:56.050 --> 00:57:00.719 James Ang: okay, that that's 1 of the challenges we're gonna have to deal with, at least for this year. 215 00:57:02.540 --> 00:57:03.170 Jeff: True. 216 00:57:03.170 --> 00:57:04.050 Paul McIntyre: That's true. 217 00:57:05.970 --> 00:57:06.665 Paul McIntyre: Okay. 218 00:57:07.360 --> 00:57:08.669 Jeff: Thanks a lot guys. 219 00:57:08.670 --> 00:57:12.659 Paul McIntyre: All right. Well, thanks, everybody, and we'll see you see you next time. 220 00:57:12.660 --> 00:57:13.270 Jeff: Alright, bye. 221 00:57:13.270 --> 00:57:13.590 James Ang: Thanks. 222 00:57:13.590 --> 00:57:14.040 Paul McIntyre: Cheers. 223 00:57:14.040 --> 00:57:14.850 James Ang: Thank you. Paul.