 And now we're going to give an even bigger, even former round of applause for Alex and Moritz who's going to talk about getting rid of nuclear weapons. Give it up, louder! Thank you, thank you. Thanks. Thanks everyone for joining us. I'm Alex and this is Moritz and as he just said this is a talk about nuclear weapons. Both Moritz and I and also our colleagues back in Princeton spent most of our time trying to get rid of them. But so far we haven't made much progress, especially lately. And 2017 was particularly difficult years. But in this business, you're in for the long haul. You have to be persistent and you want to be ready when there are new opportunities for confidence building, arms control and nuclear disarmament initiatives. Now for this talk, what's relevant is that any new initiative toward further reductions in the nuclear arsenals will have to rely on robust verification mechanisms. And as you will hopefully see today, verification will have to rely on trusted radiation measurements. After 25 years of R&D in this area, no winning technology has emerged. And ultimately it boils down to a lack of trust in the electronics that is being used for these applications. So Moritz and I thought, well, perhaps vintage computing platforms may offer a new answer to provide an answer to this challenge. And so we brought some gear and hopefully we'll demo it in a couple of minutes. But before I get to this, let me just briefly summarize where we are today with regard to nuclear weapons. There still remain about 15,000 nuclear weapons today. More than 90% of them are owned by the US and Russia. And then you see seven other nuclear weapon states in the world. Unfortunately, these numbers actually haven't come down by much over the last 15 or 20 years or so. And by any meaningful standard, it's a gigantic number. New this year was the North Korean last latest test in September, where North Korea for the very first time conducted a test, a large nuclear weapons test, 250,000 tons of TNT equivalent. And we believe this was a two-stage thermonuclear weapon. And it's significant in many ways because North Korea only tested five or six nuclear weapons before that. And it certainly now has a credible nuclear capability and also the means to deliver this weapon. We call it the peanut here. You can see this typical shape for a two-stage weapon with a primary and a secondary. Now in the interest of time, I will not walk you through the global consequences of a nuclear war. Even a limited one. And they are pretty bad as you might imagine. And personally, I believe even a single nuclear explosion in a major city would in many ways be the end of the world as we kind of know it and not in a good way. Obviously with the 200 kiloton explosion, you could wipe out an entire city in an instant. Now, lately there has been quite some loose talk about using nuclear weapons. Again, we haven't seen this in 20, 30 years or so. There is talk in Washington about a preventive nuclear war against North Korea. And that is, you know, disturbing and I'm afraid there will be, you know, more of this in 2018. And this tweet here kind of summarizes the situation quite well. You can't lose the 2020 election if there is no 2020. However, there has also been some positive developments this year. Early in the summer, 122 nations at the UN in New York negotiated a nuclear weapons ban treaty, which places nuclear weapons for the very first time in the same category as biological weapons, chemical weapons, cluster munitions and land mines. For obvious reasons, the weapon states did not participate in these negotiations, but the ideas they will join along the way. Since we're here in Germany and I'm a citizen of Germany, just as Moritz, it's worth mentioning that Germany too voted against the resolution in 2016 to start negotiations on such a treaty. I don't want to dwell on this too much. I personally believe this is a mistake. It would be the right thing to do for Germany to join this treaty and it wouldn't be particularly difficult for Germany to do so. And, you know, just to close on this, and just earlier this month, and this is, you know, it is a big deal, ICANN, the international campaign to abolish nuclear weapons, received the Nobel Peace Prize just for this reason to facilitate the negotiations of this treaty and you can see some of the key folks who have been involved and kind of made this treaty possible. So this is really one of the highlights of this year. Now, but, you know, coming back to the purpose and the topic of this talk, so what is to be verified? I already mentioned it's going to be critical to have verification mechanism in place. It's particularly important also for the ban treaty where you try to verify, you know, the number zero, and the United States, for example, has actually made the point that the ban treaty cannot be verified and for this reason it should be, you know, boycotted, which is nonsense, of course, but you get the idea people do take the verification question very seriously. Now, to kind of highlight or illustrate what involves, we have this cartoon here of a fictional nuclear weapon state, has a bunch of, you know, nuclear facilities, some of them are civilian and you may find them in other countries too, enrichment plans, reactors and so on. Some of them are military and they're highlighted here in blue and, you know, the difference, the main difference between a weapons state and a non-weapons state is that the weapons state, you know, has nuclear weapons and they move around the weapons complex, right? So what they're trying to do is confirm certain constraints on these weapons, limits on the number, you know, and the number again could be zero and they're important challenges that you have to resolve when you, you know, when you try to verify such a treaty and I highlight them here on this cartoon, we could give talks about any one of them, but today we want to talk about one specific one that's typically considered one of the maybe the most difficult one to address, which is confirming the authenticity of a nuclear weapon. So in a sense the scenario is, you know, the counterpart, the Russians, the Americans, whoever shows up and says, look, we have 100 nuclear weapons here, we want to dismantle them and we want to get credit for this, right? We want to kind of have these reductions on the books and how do you make sure that whatever they present to you is actually a nuclear weapon or a bunch of nuclear weapons? That's the challenge and that's what we're trying to solve today with vintage verification. So how do you do that? So the first thing we need to know, and it's really the only thing we need to know is that nuclear weapons, any nuclear weapon contains fissile material, nuclear explosive material, and I put the numbers here on the screen. Typically three to four kilograms of plutonium, for example, is enough or you could expect this in a nuclear weapon and that's about the amount of plutonium you would have in a nuclear warhead. I was told this is aluminum. And again, anything else really doesn't matter too much for our purposes. The key is it's plutonium or haline rich uranium, both of them are radioactive and we can use this to detect them from a distance. Now the issue is, so they have unique radiation signatures. But they are highly sensitive and they cannot be revealed to inspectors. The only exception is really the one you see here on the screen. This is a bunch of US scientists, actually some of our colleagues in Princeton involved who went to the Black Sea in 1989 and made a measurement on a Soviet at the time nuclear weapon and published the spectrum in Science Magazine. And here you go. It turns out you can actually learn quite a lot from this spectrum. So in the aftermath of this exercise, weapon states concluded, well, we cannot really do this anymore. This was kind of a bad call to go this way. So then a couple of new concepts were developed and we implemented one of them in this box over there. The first one is, okay, I can't possibly show you the spectrum that you just saw. So we have to kind of do it somewhat differently. The first idea they came up with is the attribute approach. You said, okay, you cannot look at this weapon or this component, it's in a container. But we can agree on certain attributes. So you sit down with the Russians and say, well, you know, we have plutonium, you have plutonium. That's an attribute and both sides would agree. Okay, yes, you can confirm that there's plutonium in this container. That's one attribute. Maybe there's a certain minimum mass. It has to be more than one kilogram. And we can do this also with radiation detection measurements, more than one kilogram. Certain isotopics. Even the geometry or the size of the mass you can detect with radiation detection equipment. So you make a list of attributes. One approach. We don't really like it particularly much because it's obvious how you would defeat it. Because, you know, if the threshold is set at one kilo, I just need to present, you know, 1.1 kilos, it will always pass the test. That's one way to do it. And the second one is the one we're using here for this exercise or this experiment is the so-called template approach. You do acquire one radiation spectrum from what we call a golden warhead reference item. You store it, perhaps, in some way. And then when, you know, down the road, the other side shows up with the second item, you compare the two signatures against each other. And if they match, you say, okay, I accepted the first one. I will also accept the second one. So that's the template approach. And we'll do this later in a moment. So in both cases, you do acquire sensitive information, right? I mean, the radiation detector will see everything. So what you need is a third idea, which is called an information barrier, which is really just a piece of electronic equipment that would do the data analysis and then only display kind of a red light, green light, you know, paths, fail, go, no, go result. Okay, it will not display the actual spectrum. And that's the main idea of an information barrier. And coming back to, you know, why is this, you know, such a difficult job, the problem is how can both sides, the inspector and the host team, trust the technology at the same time. Okay, so the host is worried that this machine will accidentally release the secrets, right, through some side channel or so. And the inspector's worried that the machine's actually not doing anything meaningful and will just display, you know, the result that the host wants to see, right? So this is actually something that a Russian nuclear weapons expert said when he or she was invited to a visit at a U.S. weapons lab. You know, you have some machine that, you know, does something, have no way of knowing that this is actually a genuine measurement. So that's essentially where we are today with, you know, the state of the technology. Let me just, you know, wrap up here quickly before we move into the demo part. You know, why are they so hard? I already summarized some of this. So first is, and it's very unusual for an experimental physicist or so, that you actually don't know what you're looking at, except for the fact that, you know, it's plutonium or so. So you cannot reveal or you do not want to reveal about what's actually in the box. Some information may be shared in advance, but you certainly don't want to learn anything else during the inspection. The second part is, you know, we're talking U.S., Russia, you know, China, and the rest of them. The adversary has de facto infinite resources, okay? And the adversary may be, you know, quite motivated to actually defeat or deceive your system. And just as one data point, the U.S. is currently, or refurbishing one U.S. nuclear weapon, the B-61 MOT-12, refurbishing that weapon costs more than 30 million dollars, one weapon, 30 million dollars, which is more than the weight of gold of that bomb. So a government is willing to pay 30 million dollars to refurbish one weapon. You might imagine how much resources it may throw at, you know, a system that might actually, you know, pass fake nuclear weapons as genuine. And then finally, and that's, you know, for this community in particular, the most difficult one or, you know, ironic one, the host has last ownership of the inspection system before the measurement, so they walk away and, you know, you don't know what they're doing with it, and the inspector never again has access to the system after the measurement is complete, because there may be some form of, you know, classified information left in the system. So that's essentially why these inspections are so hard and which is why we thought, you know, can we offer an alternative with vintage technologies? Now it's time for Moritz to proceed. Here you go. So as we heard, it's hard. And if you look around, who built such systems, they actually have been less than 10 built in the world, which is very few for 25 years of development. I'm going to introduce three, and then we're going to start up the computer that we have set up there. One is called Trusted Radiation Identification System, or short-trist, built in the end of the last millennium by the Sandhya National Laboratories. And this is how it looks like. It's based on the template approach, so it measures a sample and then compares this to an item you want to inspect. For this, it uses a simple sodium iodide detector. It's a radiation detector that detects gamma rays. It's very rough in the spectrum that it generates, but it's good enough for the purpose. It runs from a 12-volt battery, and it has something that's called a trusted processor in this metal enclosure. There's actually two processors and some more hardware that's installed in there. There's a green light, but it has a display and a keypad, so you control it by this keypad and you display the result on the display. And these small, little weird-looking things on the left of the picture, there's so-called eye buttons, and one of them is actually used to store the template. They're basically just, you know, small memory things you can plug into this trusted processor. So what we like about the Trist system is it's very simple in its detector setup. It's passive, and it uses this low resolution measurement information that might then be given to someone you don't want to give. And it also uses a strong tempeh indicating enclosure, so Sandia National Laboratory is actually pretty good in building these things for a variety of purposes. So this is this metal box inside of this. It's actually divided in two sides by a big metal plate. One side is called the red side, the other side is called the black side, and they only communicate through three small holes by optical means, so they're actually completely electrically separated. The black side handles the classified data, and the black side deals with the display and the keypad you saw. And they also added special tempeh boards on the sides to see if someone drilled a hole into this and make eddy current measurements before and after the measurement to see if someone has dealt with this metal enclosure. And the last thing we like, probably not the last, but the last on this slide, is it's actually pretty fast. So you can measure a template or an inspection in 30 to 60 seconds. And then you take the spectrum, condense it down to 60 numbers, and only these 60 numbers are used to compare it to another item by a simple statistical test. There are also some things that we don't like. So we don't like that it actually uses very complex hardware, even if this is only in a 586, an AMD K5 processor inside on a PC 104 form factor board, it still has 4 million transistors that you would need to verify to make sure it actually does what it's supposed to do. It also uses an FPGA for the data analysis, and there's actually two of these processor boards inside, one on the red side, one on the black side. Also, we don't really like that it's a system built only by a nuclear weapon state in the United States, and its main focus was to ensure that no information is leaked. But they actually also said that this was their main focus, so they had a hard time kind of proving to the side they wanted to inspect, that the system actually not just only shows a green light or a good sign on the display when it's asked to. There's a second system that's more recent, built actually by a nuclear weapon state, the United Kingdom, and a non-nuclear weapon state, Norway. They came together in the UK-Norway initiative for about 10 years, and they did two things. One was they simulated disarmament of nuclear weapons as a kind of life-action role-play, so they actually walked around with warheads and measurement devices, not warheads, with fake warheads and measurement devices, and tried to figure out why this is hard. But they also built an information barrier, and this is how it looks like. This is their cert iteration. They had two they built before, one by UK and one by Norway, and they had many goals to build this, so they wanted to have a simple device, as simple as possible. They wanted to use off-the-shelf hardware. They wanted to have it modular, so you see it's divided not only by geometry, but also by color into a digital board, an analog board, a low-voltage board, and a high-voltage board. They wanted also to operate from battery, so this operates from battery, so it's independent of a power supply. And they wanted it to be very robust for use, and they wanted it to be very clear while you're using it. So they came up with these three buttons, and then the middle circles are green LEDs, and the other ones are red LEDs, so there's actually green lights. But they set out to use the attribute approach. So they didn't do the template approach, but they tried to use the attribute approach, and the attributes they picked for this device are the presence of plutonium, a material that you can use in a nuclear weapon, and the ratio between two specific isotopes of plutonium. One plutonium-239 and one plutonium-240, sorry. And if this ratio was above a certain threshold, it would show a green light. What we like about this is the clear operational procedure. They had this very simple user interface. You think this is not that important, but if you have a nuclear ward in the room, you want to make sure that everyone knows what's happening, and there should be no confusion about anything, basically. They also had comprehensive documentation. Download schematics, bill of materials, and the software that ran on the machine on their website. It took them quite a couple of years, but then it's published now, and we actually learned a lot from this. And last but not least, the most important, this is actually a joint design effort between a nuclear weapon state and a state that does not have access to nuclear weapons. And they were aware of this and analyzed also what problems they had doing this. Same here, what we don't like that much. The system is based on the attribute approach, which in general is harder because you not only have to make sure that two things are equal, but you actually have to identify things from everything else. And so they use a high-purity germanium detector, similarly gamma detector, but it gives you a finer spectrum which they need, but it also requires to be cooled by nitrogen, liquid nitrogen to be very cold, and it's not easy to do in the field. They also need a very complex algorithm in the box to say, okay, there's plutonium present, or there's no plutonium present. And they have the same problem as before. They, of course, used... There is no real open chip, but they use a closed chip for the data analysis. They limited themselves also to a simple chip. They had Mega2560, an 8-bit microcontroller for data analysis. When we looked at the schematics, we also found this AT-tiny microcontroller on the so-called analog board, which surprised what this was, but they said they use it for timing issues, but just in a place where you wouldn't expect any digital analysis. This shows... This was for us the first example to say, okay, there might be something going on. Why is there a microcontroller? Why there shouldn't be? And they both have flash that probably could store images, so that might be a problem. The third example, we call information barrier experimental, or IBX, something we built two years ago. We mostly built it because it looks good, and also because we wanted something to learn, kind of a prototype to actually see what problems could we have and as a mean to teach people how you would use this and how you would build this. It's also a template approach. It can store up to three templates and then measure an item and say, you know, it's one of these three templates, and it also uses a sodium iodide crystal, a crystal and scintillator as the tris system. This is how it looks like from the inside. We try to make it very transparent and also cheap. It's cheap in quotation marks. It's less than $1,000. On the left side, there's a single board computer. On the right side, there's a custom-made board. On the right picture, you see when it measures something. And we also, in a sense, ran into the same problem. We use a complex chip. So we use this board. It's called the Rat Pitaya, which is nice because it has this fast analog input we use for data acquisition. But you can only do this if you have an FPGA, basically, because the 14-bit 125 million samples per second need to be processed fast. And if you have a proper processor and we run a full learning stack for now because it was just the simplest thing to do. So then we took a step back, as Alex said. We said, okay, what else can we do? We came up with the idea, okay, let's try if we can actually do the same with hardware that is old. And that's called vintage verification. That's the name we came up for it. Why would you do that? You take the best of all worlds and you try to build a trust through simplicity and probably also through Opsal lessons. And we use a simple detector system. Again, sodium iodide. It's widely available. It gives you the low-resolution spectrum. And then we use vintage computing platforms. We use an Apple II, in this case, which is relatively old. It has been reversed engineered even if the specs haven't been published. Not all of them. It has been reversed and quite good. So it's quasi open source. That's what I called it. And it's unlikely that someone hit a backdoor or a hidden switch in the system 40 years ago for the case that it might be used for warhead verification today. So we think this is not very likely. And another thing why we want to use this is you would like to use the BYOIB, bring your own information barrier approach. Where, you know, the host country and the inspecting country would go together and shop for hardware, basically either on eBay or go somewhere and pick a device and use this. And as the processor inside of the Apple II, there is the 6502, something we also like very much. It still has fewer transistors than there are nuclear weapons in the world. We hope to change this for the nuclear weapons. It has only 3,510 transistors. It runs usually at 1 MHz. It has 50 MHz. It runs usually at 1 MHz. It has 56 instructions. But there's an abundant number of these devices built. So the numbers vary depending on who you ask. But people say there have been billions of these chips actually made. And they have been made from 1975 and they are made more or less until today. And these are five chips that we actually had to play with. And some of them are also in these arcade machines in Alex's living room. And you see his kids testing our verification processors if they work good or bad. And they are also in the Apple II. It's the 40th anniversary this year also of the Apple II. So the computer is older than I am and that made it a fun project to develop radiation measurement cards for this. Why would we use that? Why would you use an Apple? If you look at Apple today, you think what's the benefit of doing something with an Apple? But when this was developed it was actually I think probably the last time that a hackability or the ability to extend your system won in the company policy over seamless end-to-end user experience. So when they developed this device, their first home computer that actually came in an enclosure, the two Steve Jobs and Steve Wozniak had a fight about how many extension slots this device should have. There were eight slots where people could do whatever they like. They could build cards, they could build things, and they could probably verify nuclear warheads at some point. Jobs said they should only be two. One for a printer and one for a modem, and that's enough. People won't do more with that. So we are happy that Wozniak won. And you can see this on the main board. This is the main board of an Apple IIe which is kind of a slightly more advanced version of the Apple II. In this case, I've forgotten the year. I think it's from 82. Yes. This is a board from 82. This is the board that's actually on display here on the table. And it has these eight expansion slots. And it also has some standard computer hardware. It has the 6502 processor. And it has a ROM and RAM. So we can start using it without building our own computer. In Siri, of course, you could just take the 6502 and then take some RAM and plug this together and make the same measurement, but that would be a step further than just developing cards for the Apple IIe. And with that, I think it's time to actually turn it on for the first time. I hope it works. We'll see. Can you switch the video? I don't know if the camera can see this. Turn it on. The floppy disk is not inserted. Okay. The floppy disk was missing. Okay. There we go. So first we need to ramp the high voltage. The detector typically needs on the order of a thousand volts. And Moritz is going to walk you through the details of the boards. So now the high voltage is up and we can acquire the first template. So we're fortunate. Our friends from Bochum brought us a small calibration source that we can set in front of the detector. So, you know, we can actually get a signal. So we acquire the template and here we go. Of course in the actual measurement setting you wouldn't see this spectrum. This is the thing you actually want to have classified, but we thought it's more useful to show to people here that this device is actually acquiring data. So this is a cobalt-60 source. We're not in this business. And people like cobalt-60. I mean, if you like radioactive sources, I guess, because it has these two peaks that you can see and it helps you really understand the sensitivity or the resolution of your detector. Anyway, this is going to count to the 17th count, I guess. So we go back to the main screen. If we let this run in the background, we come back to this after a while. Probably Alex actually has to push a button on the computer to acquire it. There's one issue. The detector has to warm up a little bit. So ideally it would have to wait five minutes or so to stabilize the high voltage. So we'll see how we'll be doing on the verification of the inspection in a few minutes. Okay. So what just happened? Just a very quick Physics 101. We had gamma rays coming from this source that's one type of radiation and the gamma rays hit a scintillation crystal. That's something that's this shiny blue thing. So it's basically a material that absorbs this gamma ray from the radiation source and it emits other photons but photons that are more like standard light and it emits about 38,000 photons for a single gamma ray with the energy of one MEV and these peaks are 1.2 MEV and 1. I always forget the exact number of photons that are on MEV. So all these photons get emitted in the crystal and then they hit this photo-casode which sits at the end of the crystal and they produce electrons and these electrons then get amplified by something called the photomultiplier tube which is a tube where the electrons get accelerated from the casodes towards the anode and on the way they hit these curved surfaces which are called dinodes and whenever they hit one of these they produce more electrons and when they end up at the anode the one electron from the start is 10 million electrons on the anode so it's a very, very efficient amplifier for small signals and then the anode is charged with a lot of electrons but as we saw in the demo it needs high voltage and so we designed a high voltage board first. We cheated a little bit on this here we used this CAEN or cane module which produces the high voltage for us we give this 12 volts and a set voltage from 0 and 2.6 volts and from this it makes proportional 1 kV up to 1.6 kV or 1600 volts and we didn't want to play with this voltages starting to do electronics so we used this module but we want to replace this and we build a very simple digital to analog converter which you can see in the middle based on an R2R network and this gives us this V set and it raises this over time and then we put this into later and then of course we had to build a data acquisition board a board that takes the electrons that sit on the anode feed this into this B and C connector on the left, process it I will come to that in a second and then give it to an analog to digital converter we use a 12 bit flash ADC that has luckily an 8 bit bus interface because the 65 or 2 only has 8 bit data interface and we have some control logic for these red, yellow, green and some blue ones but we decided for this time that we would not use them so in the analog part of the board on the left first you get the signal it's pre-amplified it doesn't mean it's amplified but it's something that sits in front of the amplifier and it takes the charges and converts them into a voltage pulse by a charge sensitive op amp circuit and you see it's a fast rising pulse so the next thing we do is differentiate the signal by a differentiating op amp stage and so the decay is shorter but then it gets inverted so we use an inverting stage we use an inverting stage to flip it back up and also use a trim pot to adjust the gain because depending on the build and the resistors you use you always have to adjust the signal so it actually fits in the ADC voltage range then we do some pulse shaping it's to be noticed here that actually the scope shot this is shots from our oscilloscope it takes a much wider time range and you see that the rise time of the pulse now is slower and we did this to detect this pulse easier and so fast rising we now have a relatively slow rising pulse and then we feed this in the biggest and most confusing stage which is the peak detector and hold area and it also does the ADC timing to the oscilloscope shot and this in detail the yellow line is the line of the output of number one on the previous slide of the charge collecting op amp stage the blue line this is what the peak detection hold circuit actually gives to the ADC and you see this is very nice because it rises up to the peak and then it holds the voltage of the peak and gives the ADC time to convert this and I forgot to mention it is actually proportional to the energy of the gamma ray that we detected earlier so we hold this voltage for a while and then we use the purple which is a digital signal produced by a comparator we say when it raises above a certain level then we should probably start this conversion and then we wait a little while until the signal is flat and then we emit the green digital signal to the ADC so now you can start the conversion and then if you look far to the right of the picture you see that the signal goes down again and this is actually the time when the ADC has done the conversion takes about 10 to 15 microseconds and the Apple II has read the two words from the ADC and stored them in channel in memory and also increased the total counter that counts the total number of events so it did some logic and the whole process only takes about 60 microseconds and if you remember it's a one megahertz CPU and it leaves like 40 microseconds for the processing it means it's actually 40 clock cycles and I guess if people would look at this they might even come up with a faster assembler code but that's a reasonable time and we were surprised in the end how fast we could actually sample signals before we go back to another demo just a few lessons I learned I don't know how many people in this room you know I'm a physicist I did simulations most of my time before I started doing this so I didn't build hardware and then I set out to build hardware for a machine that is older than me so how would you do that you actually start reading actual books there's some manuals online but the best thing I did for this project was to go on a bookstore and buy old books they actually cheap and they describe a lot of things that happen in this Apple II and also general things about electronics and then you have to design some things and you have to try and you have to repeat and the best thing I figured out was when I build a breadboard that doesn't work this is the breadboard where I set up the the ADC card when it doesn't work, the best is just take a new one and start again and then probably it works then and you have better ideas or you're more awake because it's the next day and last for this it's important and it keeps you motivated and it's a big problem and this could be anything from changing the heat or turning on and off the light in your room watering plants having a Apollo guidance system fly to the moon or getting rid of nuclear weapons and one of the benefits of working on getting rid of nuclear weapons it's not a benefit actually it's disadvantages the last year was bad and there were many sleepless nights worrying about what Trump and Kim Jong-un might do so when I couldn't sleep I could actually at least work on something that might be useful and if you want to try this for yourself the software for the Apple II is online there's also a repository for the hardware and if you don't happen to have an Apple II which is probably likely there's also a repository which has an emulator for the Apple II where I added special emulation for the cards we built so you can actually test out the software loading the disc that we did just on your normal Linux computer and now we go back to another demo can we switch back to the here we go so here's our template so we could either now inspect the valid item and not touch anything or if we had a second second source we could try and inspect the fake weapon let me see let's do an inspection I haven't touched the source if you watched me in the meantime I actually acquired a second template kind of doing it again hoping that the usage did stabilize in the meantime so we're now inspecting we haven't changed the warhead so to speak so we hope it'll pass but I can't promise do we have a second radioactive source in the audience let me see here we go yes we were one of our friends was kind enough to bring one along we had traveled with radioactive sources not recommended so we tried to find them in Germany they are of course completely harmless let's just wait until this one finishes I can tell you when you start calling your friends look I need a radioactive source I want to take it to Leipzig to the congress people tell you I'm sorry I can't help you with that but some do that so here we go again the code actually most of the software does the display which wouldn't be in the final version of the software so to speak right now how many bytes do we have two thousand? it's two and a half kilobytes but most of this is the display strings you see and the code that actually displays stuff so let me put the check pass that's actually pretty good so now I have the second source let me just throw it you know somewhere on the table and let's run another inspection I hit the three and well you can actually I guess if you watch the previous one carefully you can already see there's a new peak showing up on the left side so I really should fail by that by now if you have anything to add in the meantime while we're waiting we can you know do the announcement basically after this talk there will be other talks so we have to leave the stage quickly but if you want to have a closer look at the device we're going over to the CCCL and we will be at the Forum Informatiker in Frieden area which is on the first level I think and we just set it up there so if you want to have a closer look at this you're invited to come over and see we might not bring the sources just for security just bring the device just one thing in case I forget it at the very end I mean I'll wrap up in a moment but the one thing I think that definitely show with this is you the one megahertz process is actually fast enough to do this right I mean we get about 2,000 counts per second which is actually not a bad number for this type of application and that's one megahertz chip is just enough so here we go it's complete I hit the check and it failed and if you see the hex score here is pretty significant more it's the machine language coding here okay can we switch back to the main screen okay so I had I could very briefly in just two or three slides explain what the code is actually doing which is actually the beauty of the template approach and it's inspired by Tris so here's a real world spectrum looking a little fancier than the ones you just saw but basically the same thing imagine yellow is a valid item and orange is the invalid item here we have you see there's one area where they're kind of different but otherwise they look pretty similar the first thing we do is we divide it up into a number of bins small number we pick 12 but you know there's nothing special about 12 small number and then we take the average count in these bins so it's a very very low resolution spectrum or histogram if you will of this radiation spectrum and then we just compare yellow versus orange and take the chi-square statistic for you know 12 or 11 degrees of freedom and the nice thing is you know see the equation there you know you need to subtract two numbers multiply the difference and divide by another number and you do this 12 times even the machine language you know it's pretty straight forward and easy to implement and when you do this for this particular example you get for a valid item a chi-square of around 10 and you see you know distribution you can define okay what is green what is red pass and fail and if you take the orange one you know it wasn't quite it wasn't very different but in terms of the chi-square you know it looks drastically different so and you're done you have a pass-fail algorithm pretty simple so wrapping up you know where do we go from here you know can we actually turn this into a viable device for trusted measurements you know maybe with the help of this community here so there's a bunch of things obviously we would have to do you know revising hardware clean up the code there's a few things that we actually haven't done yet subtract the background you know correct for detector drift and so on replace the high voltage module you know pretty straight forward we also probably want to package the board or the whole equipment and you know some type of series enclosure so that's robust against you know tempest attacks and so on but the one thing that is probably the most important one that kind of hinges on that particular one for this particular idea is you know can we actually prove that the processor in this Apple II the 6502 is authentic it's genuine this one is actually was pulled from the tempest machine that you saw earlier on the picture but can we prove that the processor is you know is the real deal because if we can't then we haven't really you know won a lot so the Apple II board you know wasn't the next masterpiece more it's showed it already people have done amazing things with the 6502 the visual 6502.org done these very high resolution optical images of the die built a transistor level simulation of the 6502 it's kind of destructive analysis what we hope we'll be able to do is or we ask do this in a non-destructive way not the simulation but the imaging of the chip so that you can actually confirm the architecture the original architecture of the chip because we want to use the chip afterwards obviously and we hope we can take advantage of the fact that the chip is made with you know 8 micron technology it's huge compared to what we use today it only has 3500 transistor and it's very very well understood we think there may be several options to do this mentioned the high resolution x-ray microscopy ideally if we could age date the chip or the package of the chip using some forensic techniques that would be fantastic proof of provenance of the chip or perhaps even some logic testing to confirm the original architecture of the chip again we hope we can leverage in some way the deep understanding of the 6502 you know at the transistor level for this for this purpose I think that's all we have we posted everything on vintagepreferrification.org the slides the code the hardware the hardware designs and as Moritz mentioned will be outside if you're interested and you know demo the hardware and with that I guess we're happy to take questions okay so we have some time for questions so please come closer to the microphones raise your hand and we'll take the first one from the mic number 5 one question you said that you need to verify the chip that's authentic can you not do this afterwards after you use the chip for the measurement to do the destructive analysis afterwards it's a good question I think my answer would be probably not because the damage would be already done in a way that's also why cut and choose approaches where you put 10 on the table and just pick one and inspect the others at home may not be good enough for this type of application the other side would say look the risk is after the fact you learned already and it's kind of too late so ideally we really want to do it before we actually use it for the inspection okay thank you now number one please hi thank you for the talk my question was about the you had stated that you were concerned about side channel potentially leaking some information about like the spectra what you were processing over and I was kind of wondering how you would address that 6502 CPU that's probably the question we feared from the audience we basically we haven't thought of this in detail and of course you would have many side channels in the 6502 but at least that's my layman assumption is that if you can run it from a battery in an enclosure that is RF resistant enough that the system would be simple enough that people would know all the flaws against all the ways you could actually get into the system whereas if you would take a modern chip even if it's a simple chip that many of the talks you always find out afterwards that actually there's a way to listen into what the chip is doing with the 6502 people have tried to do this for 40 years this makes it secure not a full prove of course but it makes it secure than a new chip let's go to microphone number three great talk with great toys I mean what you did is you made gamma spectroscopy available with much lower, cheaper simpler equipment than other people do you also have an idea where to get sodium iodide insulator material so that you could do different spectroscopy applications at home you can actually order sodium iodide crystals on ebay or online and they're freely available to everyone who wants them and they're not even expensive and you can get the whole set of crystal with the photomultiplier which needs to be light has to be blocked out so this is why it's silver shiny so all the light that is inside stays inside and no light from outside hits the photo caster you can buy these things actually in old on ebay for a couple of 10, 20 100 depending on where you get this from dollars or euros does that also apply to red poisoned cesium iodide I guess probably not I haven't checked this is why we like these detectors there's nothing more abundant than this type of detector you can find them on ebay and they're really affordable thank you more questions step up to the microphones and microphone 3 have you gotten in touch with was at all or have you wanted to or needed to you mean with Steve Wozniak not yet maybe after this talk I mean literally if you knew when we finished this project you know kind of just a few hours ago so to speak but now we have something that actually works and I'm not actually sure what we would ask him but I feel like it thanks more questions we still have a little bit of time maybe for one more question we can run more inspections if you want okay good someone's coming up microphone 1 so thank you very much for the talk maybe a very naive question but what is secret on the spectrum what can you learn from a spectrum it's not the receipt of Coca Cola I guess yeah I have to go back all the way to the beginning but you know I mean you could make the argument and I think it's a valid one I mean it's something especially from a non-weapon you should ask what's the big deal about your secret you know it would be so much easier and that's a true statement if you revealed the secret then we wouldn't have to protect it and it's always worth asking what exactly do we have to protect we you know a kind of a middle ground here is we believe that these 12 numbers they may be just on the edge where a weapon said might say look you know what these 12 numbers for the template for this particular weapon type we might as well put them out there and we don't have to protect the template that would be a big deal but in general the idea or the the concern would be that you as you can actually learn more or less the design of that particular weapon they are weapons so people try to defend against them so your adversary might learn exactly how you designed your weapon and may actually come up with an idea to defeat it and again I'm not subscribing to this viewpoint but that's obviously something that people would bring up and weapon states have invested billions of dollars into this technology because they think they need it and they will protect the secret but the question about revealing the secret is a very important one and you know we shouldn't just take it at face value yes we have to protect every secret that you say is a secret and I think the argument if you actually give us if you're being more transparent about certain things the verification part gets so much easier and maybe less intrusive and it could be win-win at the end of the day but of course the recipe for Coca-Cola would also be pretty important sorry never mind that was the last question we are out of time unfortunately but give a huge round of applause to Moritz and Alex from Princeton University getting rid of nuclear weapons thank you thank you so much