Anmol: Hi, everyone. Welcome to the Unthinkable Tech podcast, the go-to source for the pulse on technology shaping our future. I am your host Anmol Satija. And if you have been with us before, you know we have covered some of the most exciting trends in AI, from agentic AI, small language models to how AI is transforming software development. We have covered it all. And if you haven’t listened, go check them out now. So continuing the trend,

Today we are exploring another emerging technology powered by AI. It is the neuromorphic chips. So it is fairly a new concept. Let’s say if you have wondered what lies beyond the capabilities of traditional AI and computing architectures, neuromorphic chips might just be the answer.

Inspired by the human brain, these chips are said to revolutionize industries by making devices smarter, faster, and more energy efficient to say. So to help us decode this fascinating subject, am thrilled to welcome an expert who is at the forefront of this innovation. Joining us today is Dmitry Godovsky, the chief scientist at Polyn Technology, a company pioneering the development of neuromorphic processes. Dmitry defended his PhD in 1993 on nanocomposites and his doctor of science on polymers in 2011.

He spent time in Sweden, Germany, South Korea as visiting researcher and has worked at prestigious firms like LG Electronics and Sony. In 2019, he co-founded Paulin Technology where he was the CTO and then chief scientist followed by board observing member. So, hi Dmitry, welcome to the show. So excited to have you here.

Dmitry: Hi, I’m waiting for our talk.

What are neuromorphic chips and how do they work?

Anmol: Yes, sounds great. So Dmitry, before we dive into the deeper implications of neuromorphic chips for industries and businesses alike, so let’s start with the very basics. The term neuromorphic itself is pretty intriguing, right? It hints at something inspired by the brain. So for our audience who may be hearing about this for the first time, could you walk us through what neuromorphic chips actually are?

How do they differ from the traditional silicon chips we have relied on for decades, I Maybe both in terms of design and what they can actually do.

Dmitry: Okay, so neuromorphic chips are pretty new development. It was started to develop about like 10 years ago. And the inspiration behind the neuromorphic chips is of course the human brain. So the human brain is functioning in very complicated way and the people try to model the operation of the human brain.

How neuromorphic chips mimic the human brain?

And this neuromorphic chips is just a model one of the models possible models of the human brain so the idea is to mimic the neurons which we have in brain and axons which actually neurons generate the signal and axons are transferring the signals into the human brain. So the same way we have in the neuromorphic chips. So usually we have neurons which generate the signal and axons which transfer the signal along the path to the next neuron.

The neurons generate signals the same way as in the human brain. So there are so-called lichian fire or integratin fire neurons which mimic exactly the potential rise and decay in the neurons of the human brain. The same way with axons. So they transfer the spikes. The spikes are very sharp.

Sharp impulses of voltage which are transferred from neuron to neuron. It’s exactly the same as in human brain is processing the information. actually it is not opposite to silicon technology because now most of the neuromorphic chips are based on the silicon technology.

What we have is neurons and axons are modeled by some silicon-based elements. And in such way we have spiking or non-spiking chips. And non-spiking chips include analog chips and digital chips. So maybe the closest match to biological neurons are of course analog neurons and analog.

Neuromorphic chips, but digital chips are also mimicking the human brain operation, even so the whole processes are happening in the digital model. So we have a spiking digital model. For example, it’s one of the major players in this market with is brain scale computers incorporated. So in such digital format we have to process the spikes and the neural signals exactly as they are processed in the human brain.

Anmol: Right, that’s interesting Dimitri. So these chips don’t just process data, they replicate the way our brain handles all the information with neurons, exons, and even those spikes like signals as you mentioned. So it’s almost as if we are creating a silicon version of biology, right?

So this brings me to my next question actually. Now, neuromorphic engineering, like we discussed often, draws these intriguing parallels to the human brain, right? So what aspects of biological neural networks specially inspired the development of these chips? Was it about efficiency, adaptability, or perhaps even the way our brain processes and responds to sensory inputs?

Why neuromorphic chips are more efficient than traditional AI?

Dmitry: I think all of what you mentioned inspired the developers in the search for neuromorphic technology. I think our brains are operating very efficiently. the power of our brain is just about one watt.

In comparison with huge supercomputer architecture which tried to model the brain. So the brain is a complex, very complicated system and the researchers tried to model some functions of the brain which can be transferred to the silicon technology in order to mimic the processes which are taking part in the brain.

So, the sensor output processing is done in the peripheral nervous system and it’s also mimicked in the modern neuromorphic systems. let’s say brain is much more complicated, of course, than the models which neuromorphic systems are done.

They try to be as much precise as possible in mimicking the processes which are happening in the human brain.

Anmol: Right, right, okay. So yeah, that’s incredible, Dimitri. It’s like taking inspiration directly from nature’s most complex and efficient commuting system, that is the human brain, right? And translating all that into technology, that sounds pretty amazing. So the next thing I wonder about is the real world impact with this brain-inspired design. Neuromorphic chips clearly have unique strengths that traditional systems can’t replicate.

So let’s just talk about applications here. Like what are the some of the current and emerging areas where these chips have the potential to truly outperform the traditional systems? Where do you see like, are they making the biggest impact?

Let’s explore the real-world use cases where neuromorphic chips are making the biggest impact.

Real-world applications of neuromorphic chips

Dmitry: I think the biggest impact so far is at the applications which include edge computing and sensor signal processing. So neuromorphic chips are extremely power efficient and they possess very low latency for signal processing.

And this allows, this allows making them close to the sensor in order to process the sensor signals more efficiently. And actually, the neural networks which neuromorphic chips possess, they are better suited for sensor signal processing.

And it’s very important for many applications like we in Polin Technology, for example, we are developing the chip switch processor voice information. So like keyword spotting or voice enhancement. And it’s very important now that edge applications will work remotely without external power supply. So battery based.

And in this sense, the neuromorphic chips are perfect because there are so much energy efficient that battery can last maybe 20-30 days in comparison with traditional chips which dispose the battery in one day. in that sense, neuromorphic chips are very suited to edge computing where battery powered operation is important.

Anmol: Yeah, yeah, maybe you can share some real world examples as well. you know, during our first conversation, you gave me some very exciting examples of placing these sensors in the tires.

Dmitry: Yes, yes, yes. So for example, we in POLIN technology are developing the Intelligent Tire System.

That means that we put the accelerometer inside the car tires and our chip is also placed right behind the accelerometer. And it processes the accelerometer signal and transmits it to the central computer of the car. And processed signal just determines what road conditions are. So for example, is it icy roads, snowy roads, is it sandy road, etc.

So friction coefficient can be determined. It’s especially important for unmanned vehicle or automated vehicles because when human operates the car, he can’t check what is the road condition, but the unmanned or automatic vehicles cannot check the road condition. And it’s extremely important for them to gather the information about road.

Anmol: Yeah, right. I think those are some great and interesting applications that you just mentioned, Dimitri. But as with any groundbreaking technologies, scaling it to broader adoption requires more than just innovation, I think it needs collaboration as well. So I imagine partnership between academia, industries and startup could play a very crucial role here.

Collaboration driving neuromorphic innovation

What’s your take on this? How can these collaborations help accelerate the development and the same time adoption of neuromorphic technology?

Dmitry: I think neuromorphic technology is very complex and of course the collaboration between startups, universities and large companies play essential role. For example, there is one basic element of neuromorphic technology which is called memristor. Memristor is a resistor which can store the charge and fix the resistance value and keep it memorized.

And for development of memristors, it’s too much fundamental. even startups cannot develop memristors because you need very fundamental physics and very fundamental chemistry. And here, of course, academia brings a lot of knowledge and memristors are very popular topic for academic research and they generate new ideas and these ideas are taken by startups which try to implement these memristors into neuromorphic chips.

And of course, large corporations play essential role in uniting the whole neuromorphic community. For example, we have a very famous neuromorphic chip IBM TrueNorth, which is spread free of charge in order for community to try to generate new software and new ideas and new neural networks based on TrueNorth. The same way Intel has a neuromorphic chip which is called Loihi.

And they also almost freely distributed through their network of interested companies, startups and universities in order people to try to develop some new know-how based on their neuromorphic chips. And of course, startups play pioneering role, like for example, POLIN Technology, where I work, because

Still it’s a gap between the mentality of the people. For example, we have to explain to the people what neuromorphic technology really is. And when we come to the customer, they usually do not know what is neuromorphic technology. So we close this gap between the fundamental science and application.

Anmol; Okay, so that’s a very insightful point, Dimitri, and you guys are doing an amazing job. so yeah, the fact that we are seeing different elements like memristors and neuromorphic architectures progress at different places really highlight how multifaceted this field is and complex at the same time as mentioned by you.

But as with any technology that mimics human behavior or brain-like processes, we inevitably have to consider the broader implications, right? So with neuromorphic chips approaching a level of sophistication that mirrors cognitive functions, are there any ethical concerns or potential unintended societal consequences to say like it as it comes with every AI power technology?

So are there any consequences that we should be mindful of or? How do we balance innovation with responsibility in this space?

Ethical implications of brain-inspired computing

Dmitry: I see. It’s an interesting question. Of course, now neuromorphic chips are also used to run large language modules, LLMs or big LLMs are comparable to the human brain in the way they are reasoning and thinking. And of course, ethical concerns and societal consequences are very important in this case. And actually, neuromorphic technology can possibly make large language model as compact as human brain is. So maybe we will have huge language models in very compact form factor like billions of neurons in one small computer.

And of course, ethical consequences and ethics of artificial intelligence is very important because, for example, if you make Android with human-like brain based on neuromorphic technology, it must operate ethically because it’s operated in the real world. And here, the main, of course, the main direction is to properly train the models.

For example, large language models and the models which will come as a next step to the large language models which will open artificial general intelligence or the strong AI. So we have to properly train the models in order of them to make no harm to the people and all the ethics must be humanly inputted into the models and neural networks which we are making.

So for example, now LLMs are already trained in the ethical way. there is fine tuning which allows to protect it from harming the people, et cetera, and make complicated decisions in ethical situations. So it’s up to us to the neuromorphic systems to behave ethically. It’s up to our training procedure.

Anmol: Right, yeah, I think that’s a very thoughtful perspective, Dimitri. of course, so now let’s shift gears a bit. like you said, this is such an exciting and rapidly evolving field. And with all the advancement we have discussed today, I’m sure there’s a lot of breakthroughs that are happening behind the scenes as well.

So what recent developments or upcoming innovations in neuromorphic technology excites you the most?

Future of neuromorphic tech

Dmitry: Maybe the most interesting and breakthrough ideas are coming when we try to model the brain with organic systems. For example, polymers or small molecules or memristors based on organic systems and polymers are appearing recently.

In that way, we not only mimic the brain with silicon, but we are making organic systems, so brain-inspired organic systems which can model the brain. Of course, it’s very interesting development. And actually, I, as a researcher, also study the development of polymer neuromorphic systems. And we try to make very compact compact neuromorphic systems with billions of neurons made only from polymers and metal nanoparticles. So it’s very, it will be really a model of the real brain.

Anmol: Yeah, that is quite an interesting development, Dimitri. Of course, the development of MIMISTER-based structures is definitely exciting. And those are some truly building blocks for taking the neuromorphic technology altogether to the next level. So coming to the last part, now as we wrap up our conversation, I am sure many of our listeners are taking in all this fascinating information.

But there’s often a misconception or an overlooked aspect of any emerging technology. So if there’s one thing about neuromorphic chips that you think is often misunderstood or underappreciated to say, what would that be?

Myths about neuromorphic chips

Dmitry: I would name not one but two things. it’s a misconception that neuromorphic systems are brain-like, truly brain-like. Of course, neuromorphic systems are mimicking the brain function, but brain is hundreds or thousands of times more complicated than any neuromorphic hardware.

We try to model the brain, but we cannot reproduce the brain using neuromorphic technologies. It’s just a model which scientists have done.

in order to understand how the brain works, but it appeared to be very helpful and the technology efficient itself. It’s the first thing. And second thing, maybe many people now think that neuromorphic systems will substitute traditional computers and for name and waste architecture, but it’s not true. I think it will be kind of symbiosis between neuromorphic systems and traditional silicon chips because for different it’s extremely important to have algorithmic based computing and the neuromorphic based computing which is better suitable for neural networks.

Since in our daily life, we have both processes. So we need to calculate something or we need neural networks to process some information. I think that neuromorphic technologies and traditional von Neumann based architectures will coexist, maybe in one form factor. So we will have device which will have coprocessor, one traditional and neuromorphic one.

Anmol: Yeah, true, true that. like those were some great points, Dimitri. And I think it’s you said it rightly that it’s easy to assume that neuromorphic chips are a direct replica of the human brain. But as you have highlighted, they are really just models that mimic certain brain functions. Right. So that the complexity of human brain, I think, is still far beyond what current technology can replicate but these tips that you’re talking about are exciting step in that direction for sure.

So Dimitri, it’s been an absolute pleasure and having you on the podcast today and your insights into neuromorphic chips and their potential to transform industries have been truly eye-opening. Thank you. Thank you so much for joining in today.

Dmitry: Thank you. Thank you.

Anmol: Yeah, and to our listeners, thank you for tuning in. We hope this episode has given you a better understanding of how neuromorphic technology could impact the future of computing and broader tech landscape. So stay tuned for more in-depth conversations with the brightest minds in technology right here on the Unthinkable Tech podcast. Until next time, I’m Anmol Satija and we’ll catch you in the next episode.