The FUTURE of Computing IS HERE - Photonic Chips

what will the computer of the future look like for the past few decades we have seen conventional computing meet the demands of everyday applications it is true that silicon transistors have only gotten smaller over the years however a single silicon atom is 0.3 nanometers so there is definitely a limitation to how small a transistor can get concurrently these types of chips also have severe drawbacks when it comes to ai modeling 10 years ago the biggest models were around 10 million parameters and could easily be trained in a few days on a single gpu today there are some models that are over 10 trillion parameters and in turn taking thousands of machines to train things like automated cars speech recognition and even ar assistance are highly dependent on machine learning so there is a big drive to develop a different type of computing architecture one of the front runners is photonic computing there are multiple startup companies looking at different problems in which these chips can solve but generally there are magnitude improvements in terms of speed and energy consumption one notable chip is the pace by light intelligence it has ten thousand photonic devices on one chip running at one gigahertz the core of the pace is an integrated silicon photonic chip which combines digital and analog signals photo detectors turn optical intensities into electric current and this entire configuration is mounted on a pci board so it's not entirely photonic but this is a stepping stone which will likely be taken by many startup companies and this means that the chip can also be built through existing semiconductor fabrication techniques ultimately the pace can tackle max cut problems 100 times faster than a high end gpu the max cut is one of the simplest graph partitioning problems to conceptualize yet it's one of the most difficult optimization problems to solve applications include network designs circuit layouts and even data clustering another really neat development is light matters photonic chip so there's a lot of skepticism on how many optical components you can integrate on a circuit so they're developing a chip which can handle different wavelengths at the same time without losing coherence it would be interesting to see how many different wavelengths can be simultaneously running in the same hardware but this could be an answer to the scalability problem their chip will eventually act as a central unit which will power an autonomous ev another really infamous computer we hear a lot about is the quantum computer but the fact still remains that qubits are extremely hard to isolate decoherence is inevitable and errors eventually creep in having said that it basically is meaningless to have all these q bits without proper air correction we do have quantum annealers such as d-wave which can solve optimization problems but you always need to translate it into a cubo model so it's a little bit problematic where a real quantum computer can implement any quantum algorithm keeping this in mind there are some companies which claim to be able to tackle this quantum problem by utilizing squeeze states of light and it's basically a programmable system of entangled photons the chip turns photons into squeeze states after the interferometer performs operations on the squeeze states the photons leave the chip where external detectors perform the readout for now the photon detectors have to be cryogenically cooled but the company claims that they have the framework for million qubit machines in the end we do have a few photonic chips which are basically an experimental phase but we still have a big problem when it comes to storage as of right now we do have ssds which can read and write in gigabytes per second but they also have strong limitations in terms of capacities dna storage could be a possible alternative as it can be encoded as four bases instead of zero or one in theory dna drives could produce exabyte scales or one billion gigabytes but the problem is is that these are very difficult to scale up and control however there is research being done on this holy grail of storage as georgia tech has made a feasible way in organizing dna strands for storage they have achieved this by offering a one inch chip with several microwells strands are grown in parallel inside these wells at some point these chips will also include a second layer of electronic controls made in your standard complementary metal oxide semiconductor that will initiate the chemical process of building each strand of dna one base at a time the challenge is to push the density of the dna as high as it can go and ultimately adding electrical control there's still a lot of questions on access time bandwidth and volatility of these chips in conclusion we will eventually need a different type of computing architecture to meet the demands of machine learning photonics could very well be the answer to this problem if we can solve the initial fabrication issues along with figuring out how to increase processing speeds without causing decoherence in wavelengths but if we do figure out these problems this type of computing architecture will be a very strong framework for future machine learning and may even lead to something which can imitate the human mind anyways i would like to know what you think about all this so please leave a comment like the video if you enjoyed it and also make sure to subscribe to my channel

Channel: Tech Planet

Views: 41,939

Keywords: future computer, quantum computers, light computer, fastest computer, qauntum computer

Id: Xhn9vw8ur0A

Length: 5min 37sec (337 seconds)

Published: Sun Mar 27 2022