Neuromorphic Engineering — Meet the Future of Computers
Although progress in developing neural networks is speeding, the hardware that these programs and applications work on is hardly changing. A lot of companies, including Neuroseed, have already faced the challenge and are rethinking current processor architecture to fit the needs of artificial intelligence. A completely new approach is proposed — to remake chips so that they function more like human brain. This direction is called ‘neuromorphic engineering’ or ‘neuromorphic computing’ and this is the topic of our today’s article.
Let’s start with understanding the difference between contemporary processors and neuromorphic ones. Firstly, neuromorphic semiconductors imitate bio-neural architectures existing in people’s nervous system, so they mimic the structure of our brain. Secondly, digital transistors in current chips are laid out in traditional lines whereas billions of transistors in neuromorphic semiconductors act in parallel via inter-neuron connections, which are similar to synapses in the human nervous system. So, such a type of chips can process data in an analog fashion, not digital, varying the strength of the signals while sending information. This suggests that a lot more data can be put into one jolt. Yet, there is one difficulty — to be able to accurately control these analog signals. No doubt, the intensity should change, but in a regular controlled way.
When at Neuroseed we reason how supercomputers might copy our brain functions, we come to the idea that they should be capable of immediate learning and processing tremendous amounts of information in order to create human-like ‘implanted memory’. These requirements can be fulfilled if supported by neuromorphic engineering and so revolutionizing lots of aspects of our life — traveling, shopping or everyday routine.
Up to date, in quite a few experiments neuromorphic chips have proved that they can successfully process voice and imagery. They are of really great use when it comes to excessive processing, because they can perform it much more easily and rapidly than current chips. As for companies which are now mostly investing in looking into and developing the neuromorphic environment, they are IBM, NeuroMem and Intel as well as Qualcomm and Google.
There is still a long way to cover before we can be sure that neuromorphic processors are suitable for massive manufacturing and completing real-world tasks. But today’s researchers, scientists and developers are moving towards a big dream and colossal leap — to completely alter the way computers think — and it requires a lot of research and work. With a team of committed and constantly self-improving professionals, Neuroseed is ready to play its role in the trip to a ground-breaking future.
