Neuromorphic Computing, AI Chips Emulating the Brain with Kelsey Scharnhorst on MIND & MACHINE

hi I'm August welcome to the mind and machine show where we explore ideas on the future today we explore neuromorphic computing with computer chips that emulate the biological neurons and synapses in the brain neurobiological chip architecture enables machines to solve different kinds of problems than traditional computers the kinds of problems we previously thought only humans could tackle my guest today is Kelsey Scharnhorst Kelsey is an artificial neural network researcher at UCLA her research lab is focused on creating neuromorphic computer tips and further developing their capabilities we'll talk with Kelsey about how neuromorphic computing is different how neurobiological computer architecture works and how it will be used in the future let's dive in so Kelsey I am really excited to have you on the show great to be here you're an expert in artificial intelligence and you're doing some really powerful work that we're going to get into very quickly within artificial intelligence there's sort of a software side and that's what we hear most about and then there's a hardware side and you've chosen to focus on the hardware side what drew you to hardware as opposed to software in addressing AI problems so I actually did my undergrad in chemistry and worked in solar power and when applying to grad schools I was applying to all environmental science projects and this one project pqube my interest and it was the one that I ended up going with because it was the most interesting to me okay and you jump from chemistry yeah into computer chips yes that's a big jump the fabrication of microchips is all chemistry based and as they get into these chips in particular yeah are very chemistry based okay there's been a evolution within the hardware side of computing based on Moore's law right this is this exponential increase in power every two years or so a doubling of power having of the amount of space you to put transistors into right yeah so why is that coming to an end how is that slowing down so we're starting to get to the fundamental limits of this law so you can only pack in circuits so close to each other and our iPhones today are 15 nanometers and they have since made eight nanometers but to get to 4 nanometers you are essentially pushing the limits of light oh and this is how we do patterns of tiny circuits we're maxing it there's really no clever way to get around this yeah and it is physical properties it's gonna become incredibly expensive I think we can get past 4 nanometers but it would be not cost-effective especially on on large scales okay and you were also telling me about a limitation a bottleneck that we're hitting on the traditional approach to making computer chips yes so the computer architectures in computers that we use today they have a separation between memory and processing and in between pulling information from the memory storage of the computer to doing processing there is this bottleneck so how fast you can go between these two areas determines how fast you can compute basically and where we're getting to the point where that is a problem if we want to start increasing the complexity of the problems that we're posing to a computer so is it the fact that the chip in the processing is separate from other parts of the function the memory and all that that's the that space difference is the bottleneck yeah the ability for it to travel down that gap yeah and I think there was something in the news recently with them well maybe I won't name the company but there was a chip that had that very close so their chip was really fast but it ended up causing a security breach oh okay yes so yeah okay and I asked all these questions to set up the big topic today which is that you are building neuromorphic chips which is establishing this thing called neuromorphic computing mm-hmm and essentially these are a type of smart chip that differ from the traditional path that has these issues we've been talking about what are neuromorphic chips and how does that you know how does neuromorphic computing changed the game so if you go to the root of the word neuromorphic the morphology or the structure of these chips is based off of neurons so we're taking biological inspiration and then applying it to an inorganic system that we could use for computing so it was the idea was first invented by Carver Mead who founded Cal Tech's computer science department and he also is a one that coined the term Moore's Law so Gordon Moore came up with Moore's law but Carver made his good friend ended up calling it Moore's law a highly influential guy yes it is in the middle of everything at this time and when was this 70s 70s early 70s yeah and so the chips are inspired by neurons and synapses in the way that they interact yes so if you look at mammalian brains a way that we process information we take tons of multi sensory inputs visual hearing smell sight touch all of these things and we're processing a lot of the time without even consciously thinking about it we have background activity but most importantly our brains are very low-energy compared to a standard microchip so it should take a lot of power just to do their basic processing yeah and that's the heat issue and we always hear about with computers yeah the pictures of these are astonishing to me and this is what really blew my mind when we first started talking these look like neurons and synapses in the brain they look completely biological yeah so that was our goal you can you describe them for people listening who are watching visually what these these neuromorphic chips so yeah look on the inside they essentially look like a densely interconnected plate of spaghetti no I mean it looks stiff they are stiff and we do so we've done a lot of work over the years on this project in optimizing the density of the nanowires so the way that we do that is we start with standard microchip fabrication techniques with photolithography is what computer chips or any microchips and electrical devices are made from today so all you do is you coat a silicon wafer with some sort of polymer and shine UV light on it and depending on what pattern this light has been shown through a mask then you can get these highly patterned electrodes or whatever you're putting down and following that so we have a basis of very controlled and ordered system and we end up putting copper posts down and this is a specific to this process but you can use different metals for different forms of this technology but you have a copper grid essentially tiny little posts and these are the seed sites that you would then put in a solution it's a seed that is gonna from which everything is going to grow around yeah yeah so it's not biological it's completely synthetic completely synthetic and follows kind of the laws of thermodynamics where it's taking this path based on the amount of ions in solution they look beautiful when they're when they're done well yeah just for those listening there are like bulbs in you know scattered throughout space but these either rods or tubes like connecting them I mean really what you think of as neurons and synapses in the brain it's as nanotechnology so this is microscopic super super tiny but when you blow it up and look at it it's got these bulbs which are the neurons and these connectors narrow thin connectors between them and it's this network in 3d it just goes deep and it goes wide yeah yeah so in every single network that we grow is completely different than the last so every single one is very unique much in the way that humans brain structures would be the important thing is that you have short and long connections okay and these can lead to short-term long-term memory and distribute an activity throughout the network the memory is combined with the processing unit this is before we had the bottleneck of the memory in the processing unit separate in the gap between them yes literally the two are the same thing yeah what makes the short-term memory versus the long-term memory potentially everywhere that we have a nanowire that crosses itself all right well another one any overlapping wires they can have a junction and where these are crossing you would have an insulating layer in between them and from the silver or the metal you build up atom by atom to form a filament between these two but we have essentially a billion per square centimeter of these connections so sweet a billion note like bulbous nodes and little thin connectors yeah Wow we we had one grad student so the data just moves through it is there data transferring through it or how does that analogy of data being stored on a magnetic hard drive connector that relate to these biologically inspired networks so this would be more of a physical training in terms of memory so so they're no ones in zeros here no one's in zero that's huge that's really different yeah so this is it's a what we call analog computing so this is straight-up analog yeah so it's memories like muscle memory in a way yeah so instead of a zero or a one and on or off we have we have off we have on but the if it's the short-term memory it's kind of a thin filament and will dissolve quickly if there's no reinforced electrical impulse and that's like trying to retry to remember if we don't exactly have repetition yeah it fades mm-hmm just like literally the exact same thing yeah and then so then when you have that repetition you get a thicker filament that forms and that is more robust and will have a longer lifetime for viewers and listeners who don't know how the brain works and this is my understanding correct me if I'm wrong but the way the brain works is you have these sort of connections between neurons and the things you study and practice the connections is grow like literally grow stronger and your ability your ability to be skilled at something or knowledgeable about something is to have strong connections between these sections of neurons and to be ignorant about something is to have weakened or no connection to weak connections and that when you learn you build these connections I would definitely agree with that so that's really how the brain works is literally how these neuromorphic chips work see we're replicating it's not inspired by it is the same way that brains work is that fair well brains are extremely complicated so I don't want to say it 100% that we're copying the brain but we are trying to model the way that the brain performs computations because it's much better at complex problems so anything that seems difficult to predict is something that we would be looking at with this sort of system just before we move on behind the the physicality of these it again is it starts with a seed in what's firs the original network of growth because you know when you're learning it's that's fostering growth but you need to have a certain level to be able to interact or have that initial processing the way that they grow from the seed we put them in a solution that ends up having the metal ion we want to grow and it does kind of a dosey doe effect for the seed so the seed swaps individual atoms with atoms in the solution and from there it starts to so it replaces the entire post and then starts to grow out with the solution a kind of knowledge then because it's taking its building pathways from the solution yeah I don't think I would I think that it is following the path of least resistance or you know entropy you look at but it has a motivation to expand that there's some inherent tension that wants to reach out it's taken a while to get the current concentration so it has to do with this wave front of ions so if you have a post and all of the ions in the front are eaten up too quickly then it'll start to branch out so we started making very dendritic structures that were branched and they were less reliable than when we had the longer pretty nice-looking okay no wires that organization helps yes the functionality yeah and I would definitely say this is a self self-organized criticality within the process of its growth okay and from there we we add like kind of a cladding to them with sulfur so it's just a gas that we and that is what allows it to switch because without that layer nothing visualizing this thing growing into this very organic like thing so if so these chips are a new kind of processing but basically there's still computer processing though they work you know in an analog form like the human brain so do they process similar kinds of information with similar kinds of outputs as traditional computers you know like it's like a powerful calculator or are they more like the brain and that they're learning there our making mistakes they're trying things they're expanding as they grow up I mean is there a sense of growth and mature a Shinto them they're gonna do completely different things than computers today so that's that's important to note because computers today are very good at what they do they are very good at brute force calculation if you want to do a mathematical formula and type it into your computer and it's very good at that but if you wanted it to say look at an image and try and figure out what was in the image you know you wouldn't be able to do that on your computer unless you were using machine learning which is typically a soft word yeah so that's very software but in something like this you could take that image you'd have to transform it into electrical impulses to give you know send it into this system its chip and then from there ideally you would have a training period where you're setting setting the weights of the outputs because we have multiple outputs for each chips we have different generations of the chips where we have four electrodes 16 electrodes we would have the training program for that for it to take in data sets so study something it would be to set the weight so the idea is to ask it a question and you want one answer but you get 16 outputs so how do you get one answer you combine them averaging well you could average them but that would be the weights would be all the same but you could do a little better where you could say like oh you know number one is pretty good I'm gonna give that like 0.9 so multiply that by 0.9 and then like you know if it were a two electrode output you could say number two let's times point one so there's a kind of a trauma patient yeah involved so the user needs expertise yeah this is something that we would have the computer huh determine the computer would have sort of interpret itself yeah you got a different computer different it would be one of the the standard computers okay so standard a meter with probably some kind of AI software would be evaluating the results of the neuromorphic computer yep Wow yeah so they'd be like friends that help each other exactly so it would be something that you could add on to the standard computer architecture and then it would only be with complex problems that you would want to use so neuromorphic chip because there is a in allowing it to learn on its own and determine things on its own you do have to give it a level of error so you can set that error to what you want a lot of people like the plus or minus 5% maybe plus or minus two and a half percent so you said the acceptable range of error and what limitations does that impose on the chip with a computer in terms of output ah wouldn't you always want a smaller range of error if given a choice ideally but in things like if you're trying to predict we've done traffic prediction which is really turns out incredibly hard to predict but we've we've done it and I had fairly decent results you wouldn't want to say 100% that this is how it's gonna be because you know any human could say like of course you don't know that to give you an intelligent guidance mm-hmm but not a definitive yeah answer okay and initially it sounds like these would be built into traditional machines right you wouldn't necessarily need to have two machines yeah you'd have either a chip housing with both built in or at least a computer frame yeah I think Piru casing that would be probably the way to start it's there's a lot of potential for robot brains um which it's certainly pretty far out projection but they because these chips utilize such small amounts of energy right yeah yes we need some energy right yeah robots these days um a lot of the times they will send out data to remote the cloud yeah the cloud so somewhere wherever the the machine learning code that's been designed for them to recognize trash or whatever they're doing to do the heavy processing that requires a lot of power yeah and then and then it pings it back to say yes this is trash you should pick it up or no this is someone's swimsuit leave it alone I'll leave that day with these neuromorphic chips can do the whole thing they're not reliant on the traditional architecture I think it's very possible we don't there's no way of knowing is that the hope as you guys build these is this a goal I think in 30 years it's it's possible okay yeah and what what other examples of things are well-suited for this kind of processing and we're starting to get really good at pattern recognition with the software AI when they have very large deep learning databases like tensorflow there's pretty high accuracy but then if you then take that and try and ask a computer in what context you're seeing an image and they're they're not gonna be able to say like oh this is a kid's birthday party or oh this person's being robbed like they just see images and are able to identify with images are like that's a gun that's a birthday cake that's sweet spot of that type of thing but it can do that type of recognition better than anything else we know about is that is this the leading way to do that kind of process all that kind of problem or is the hope that that it will be yeah I think that the hook is there and I mean there are other forms so we have a very unique structure most people even if they're doing more than neuromorphic computing they will have still very ordered systems so you see a lot of crossbar arrays and so these are things that you know right so ordered and you know how many connections there are our system is not ordered I would say the structures my tank yeah with all the the rapid development of software-based AI and machine learning in particular it seems this kind of processing chip would pair really nicely with machine learning software I think so too yeah absolutely there's plenty of potential there right how would they work together I mean how would the software tap this unique architecture within the architecture because of the massive amount of connections we end up getting these feedback loops within the device without even really meaning to but this is this is great because this is what a software program engineer would be writing in their code so they have the intro and then they have this sort of black box where it goes around and around and it's doing some processing you don't really know what it's doing and then you get some output and that's on a software side and then we have hardware that does essentially just that so if we were to use this hardware to do that process it would be much faster then when your computer does it because when the computer is learning sometimes those sorts of programs can run overnight even depending what they're doing but the real-time processing of a micro chip like this I think that we could speed up the machine learning pretty quickly even with interpreting the outputs that that is the biggest challenge with these sort of microchips is deciding how to encode the input and then deciding what the output is gonna mean okay so interpretation is a big part of it yeah but if we were to start devising machine learning algorithms that are you know based on these devices and we can get better and better results okay wow that's it's just astonishing to me you use the phrase that these computers allow for error which is what you're talking about earlier that's good little deeper into that what are the advantages of allowing for error and then how do you deal with the I mean the obvious downside is that what does it mean if it's if it's spitting out errors but let's start with what's the or the advantages there's certain kinds of problems this can tackle that others yeah I think that that's a main advantage is that it can tackle very common I keep using the word complex problem but it it means that it's not yes or no answer it's not clear-cut and you want to be able to to solve it quickly is it subjective yeah good it could definitely be a subjective question interesting so it's got opinions in a way it might give you a percentage of 40% oh okay okay but I might be hood okay so it can I can do calculations on probabilities and give its recommendation and judgment based on the probabilities yeah one of the larger markets for this is stock market prediction good magic so pretty big market and everyone is very interested in getting those predictions even a fraction of a percent I mean millions of dollars okay so yeah and you said the distributed activity enables power laws in this context what does that mean for so what is the distributed activity okay so we have these complex pathways in the device so if you input a signal in the upper corner of the device it's gonna kind of go around in this like very unintuitive way based on how these switches are designed at the time which you know they move and they change so instead of going from point A to point B like in a straight line it's gonna take these interesting pathways and then it transforms the information so we always say that it's a nonlinear transformation of information and that is essentially what people are doing with machine learning as well so you transform this information and then we get it we pull it out and then we add it together so the way that it's working though is it's all throughout the device is performing some transformation it distributed throughout the device or throughout the chip throughout the chip okay so the chip is distributed okay and then the power how does that translate into a power law so if you look at that activity and plot it on this thing called well a power spectral density apply it's you know it doesn't matter if you understand what that is but it just means that if you put on a log-log plot you will get a straight line and the interesting thing about that is that human brains produce the same power spectral density plot it's a straight line and they even have shown in multiple papers you know you're a sleep stay awake state have different slopes they represent your ability to solve problems and to think is that what the curve is that the straight line is representing I would say it's more just based on brain activity it's not necessarily intellect okay because everyone has them and right we can't it's some kind of processing power gauge it's yeah it's basically a current flow an electrical current electrical impulses yeah so that's in higher is better yes a higher slope is is more activity and yeah when you're on drugs you have these chips how do they produce a different kind of curve or line and so we essentially just manipulate the current flow Oh control of the line and yeah so we have changed that line oh so you've control over the line yeah which is not the case in a human brain right I'm assuming I mean short of taking it yeah yeah I mean you're more just observing in it with this we don't have it's not that we have a hundred percent like control like we can't set that Loeb and then have the current flow meet that not yet anyways I think that's very possible but right now it's it's we set the current and then we kind of observe and it's basically like this and then if we change it we can change it you know each chip is different so these slopes that we are changing from can vary is from chip to chip that's fascinating yeah and so this whole process we've been talking about this is a form of neural networks right maybe we heard about neural networks I just wanted to connect the term and but a lot of times when we hear about neural networks it's sort of a software based neural network yeah yes this is a hardware neural network advantages/disadvantages or they do they is it best to put the two together and they could certainly work well together yeah we we talked about natural computing a lot so computing inspired by nature utilizing structures inspired by nature so certainly this is explicitly many some ways replicating nature if you're running software on a neuromorphic chip if you're running digital software on a analog chip how does that work that just occurred to me oh you have a different kind of software to run an analog chip so we have a pretty expensive set up hundreds of thousands of dollars in the room that I work in and if we have that so that we can make sure that all of our signals are true and we believe what we're getting and then from there that is actually a computer so our it's by National Instruments so it has a bunch of little chess ease that you can plug into this large box is the software running on the traditional chip and then certain problems are sent out to the analog neuromorphic chip we what we would do is just send either a voltage or current into the chip and then we would look at all of the outputs and then all of the competing that we do we do post-processing right now but certainly in the future we could be doing real-time we have we have some programs that do some real time stuff but they're not very good right now but so we take all of the outputs and then we do that sort of thing where I was talking about earlier with the small bit of training to set the weights of each output and then we combine those and then we're getting our information okay so there you have a traditional computer running traditional software see this obviously doesn't work without a traditional chip writing software then we're just trying to figure out how the software that interacts with the neuromorphic chip is it just that the traditional chip is sending out certain requests to the neuromorphic chip and it's translating between the software and the neuromorphic chip or does the software communicating directly with the neuromorphic chip well we have a user interface so it's really us it's that's inner like we're deciding what we're sending to the the neuromorphic chip is your interface on a computer screen yeah yeah so the software which is written with ones and zeros I assume can interact with an analog chip we use or am i opening a Pandora's box of complication that background to understand no no it's actually pretty simple so we use LabVIEW so this is a graphical programming language so instead of line by line code you have these pictures with wires and that's what we program our inputs and our hole we call it a it's a virtual instrument to send and receive signals so from that I can decide hey I want to put a square wave in and if you if we put a square wave in we could call that you know digital if we wanted to but I mean the way that we're doing it I would still call it an analogue chip but we we also have the ability to send sine waves different frequencies multiple inputs now it seems to me that this kind of explicitly analog but very similar sort of neuro biologically inspired process if computers and machines are ever going to achieve consciousness this seems like the path to that did that seem plausible to you from your perspective yes absolutely it's something I think that sometimes and it's certainly a little terrifying whenever I think about it but it's typically not brought up and look in this area senior colleagues or not we don't talk about it philosophy of consciousness and and how this connects I think it's it's something that because we don't know they don't want to you know write breach Pandora's box but I certainly think that you would not be able to find consciousness and a machine without a structure that has been through evolution and because that's an interesting point so this had a form of evolution in its creation yeah that's fair to say the way it it grew they literally grew the networks or is the evolution I would drink thinning and in development of the connectors that's that's one way to look at it I would think more along the lines of you know there's no threat by nature for its survival so the only way that we could create consciousness in a machine is if we were modeling it off of something that has been through that process which is the mammalian brain suppose W that the architecture we know can lead to consciousness right increases the odds at least if we had a conscious machine I don't think that it would necessarily decide or well start making decisions this is something that unless it was inherently programmed to do what it would have just spontaneously evolved to start making a decision when it was it's never been part of it and it has achieved consciousness only by seeing but if you know but it works the question to me is just that if it's strengthening these pathways and these neural networks are strengthening and learning through behaviors activities pattern recognitions actually large quantities of pattern evaluation I mean we don't we don't know what creates consciousness we don't know what the link is between the mechanics of the body and the experience the the sense of being conscious yeah but it seems like an element of that must be these growth pathways in our brains the neural pathways it just seems like a fair guess the best guess we can perhaps make and so my understanding is these work sort of like deep learning processes right so we don't know you can't after it spits out an answer we don't actually know how I came to that answer is that yeah correct yeah so in some software's like that and these chips are like that you'd have to imagine that since we don't even know how our own processes led to our consciousness that this ability to sort of run especially in an analogue context and develop and come out with answers we can't anticipate would be a path to consciousness I think it's very possible what should we be on the lookout in this field as next milestones what's the what's the thing you guys are trying to achieve next in this field certainly this field is going to want to go public pretty soon putting them into product products yeah so IBM has one one chip that uh TrueNorth IBM's actively working on this yeah they've got one there's a few others Qualcomm head stuff in your lab at UCLA is working with industry are you working purely in an academic context for the sake of understanding purely academic at this point but the idea of the academic breakthrough is I would assume you correct me if I'm wrong is to inform industry in practice right so do you publish your work and make it publicly available to scientists at other reach research labs yeah okay yeah and we have collaborators all over the world we've worked pretty closely with the reservoir lab in Belgium and it's a global yeah Portland State University to UM do a lot of our machine learning for us nice mm-hmm awesome well I mean this is a mind expanding to me is I can't even I'm still wrestling with the implications of it but it's also exciting that it takes us off the track that we've been on since the advent of computers and puts us on a new track and the fact that it's analog and that it is maybe not literally biological but it's in every way functioning biologically it's just massively intriguing to me so thank you so much for sharing all this where can people find out what you're up to and follow your work I can visit my lamps website at UCLA search and rescue um you will link to that below okay and my LinkedIn it's so amazing I am like again I've got lots to digest and that's what I love about these conversations so thank you thanks thank you guys for tuning in if you're watching on youtube and you like this kind of conversation hit the subscribe button and feel free to leave any comments below I'll be active with anything that pops up down there if you're listening to the podcast ratings and reviews are super helpful helps other people find the show and until next time thank you

31 thoughts on “Neuromorphic Computing, AI Chips Emulating the Brain with Kelsey Scharnhorst on MIND & MACHINE

  1. Great editing for a mind boggling interview. How you kept up with each other is truly impressive. My area of expertise is the quantum physics of semiconductors and her ability to explain things is simply staggering. (in real time, I mean, versus writing out a proof or explanation!)

  2. A scientific stance doubtless and not too difficult presented so it was comprehensible in my opinion. People could be beautiful there is no doubt, science has to be earned we can doubt it but everyone has to go this pat, there is no other way to be a scientist.
    Surely, technic and/or technology and science are powerful, but it shouldn’t contradict to the natural ongoing developments, because they are deep in space and time and who knows what else. In the XXI century we just learned a little bit what is going on inside atoms, this is the end, probably not. I hope this comment is a greeting to you both and to the presentation what is going on eventually.

  3. Best application is when you put NM together with ML … like an accelerator or coprocessor. Oddly enough best application is AI processing chips like Google's TPU could offload some task to it or even the NM processor becomes a teaching accelerator while the TPU remains a learning accelerator or vice versa … (power dependent).

  4. For power I would say NM is low power because firstly it's inherently asynchronous, secondly it's concurrent or massively parallel, third memory is localized to processing so less distance processed data actually moves, the data encoding is inherently low entropy (think gray code) so it switches with less power, and fourth is it should hopefully take advantage of Moore's Law or whatever is left of it.

  5. Interesting approach although I believe the electrical engineering approach is probably the preferred method … for example in the VLSI design it would be highly ordered and structured with modular components that look like ordinary processors. Each NM processor would be equivalent only difference is the teaching algorithm that is run across that volume based production network. There are some basic Analog circuits like the Schmitt Trigger designed to mimic natural phenomena; so in essence the circuitry building blocks already exist to make such switching networks and synapses.

  6. seriously she looks like and behaves like the girlfriend of Sherlock in elementary 😛 (bonus: she was also a computer wizard)

  7. Sounds absolutely amazing, but I really don't get how you program these things. From what I gathered they aren't purpose built and have general computing capabilities though. Mind-blowing stuff, when you're used to thinking of computing in the framework of the traditional van Neuman architecture.

  8. WOW just WOW, so let's suppose we find a way to "download" information from a biological brain, that means we can transfer our consciousness to these analog hardwares (Neuromorphic chips)? … if so, can we say we could be immortal (our consciousness)?

  9. Still the best and easiest way to create counciousness is to have sex with someone…

  10. Did you experts just literally say you DONT KNOW HOW THE 'ROBOT' EVEN IS GETTING ALL OF THIS 'INTELLIGENCE'???? WOW

  11. what? Do we already how brains work? That's the biggest misconception. In an addition, the neuronal network is infinitely more complicated than the artificial neuromorphic computing this videos describes. There are maintainances, repairs, growth, etc with biology, and these processes are realized by molecules mostly by proteins.

  12. Interesting Issac Newton was doing this using crystals in his Lab when he got deep into Alchemy….I imagine future AI will use a combination of Organic Neuromorphic Networks ,Advanced Traditional Nueral Networks, Qbit Chips ,Blockchain, and Emerging Agents Interfaced into one system, probly tied into a Photonic Information Cloud…with Information stored on a type of synthisized DNA.

Leave a Reply

Your email address will not be published. Required fields are marked *