Search :

Hybrid Quantum Computers

The development of practical, affordable room temperature hybrid quantum computers [HQC's] will change the machine learning world as we know it. What better reason to try and build it?

To that aim HQC is a fundamental part of our strategic development program. First by the development of software systems able to exploit such a device and secondly the development of the processors themselves.

In 2014 Toridion developed what is most likely the world's first fault tolerant quantum neural network. Based on a blend of probability driven memory architectures and a proprietary integrated qubit simulation the Toridion Quantum Neural Network [TQNN] has several unique capabilities that elude it's classical sub ordinates such as a high degree of Neuroplasticity and the ability to generate genuinely new concepts from error conditions that avalanche toward optimum solutions. Combined with quantum tunnelling simulation optimisations, even a modest system can perform truly remarkable tasks. This technology is now making its way to full blow API stage.

Why software first?

We want people to use it! That means being sure that the technology is presented in a way that programmers and coders can use it. Before we could do that it was important for us to understand the types of problems that QNN technology could solve. At first the slow speed was a factor, but over the years optimisations have resulted in QNN software being deployed in realtime video processing and numerous other practical applications. In short, the confidence is there that QNN's have the ability to learn, interpret and classify data just as well as their classical counterparts. The difference is that QNN's when deployed correctly can outperform much large classical systems of far greater cost and complexity.

What is hybrid quantum computing?

Most stories you read about quantum computers talk of hugely expensive, large and cumbersome machines, additionally they often talk about super cooling of the chips using enormous cryogenic units. That's great for science or if you are NASA, but if you want one in your bedroom – that's just not going to happen unless you are Tony Stark.

Hybrid is a different approach completely. Here we are looking to the software (which we now know provides significant advantages over existing systems and comparable capabilities with regard to machine learning) for clues. Knowing the software can deliver is crucial. We are now (and have been for some time) reverse engineering the core components of the software technologies and building new Application Specific Quantum Integrated Circuit [ASQIC] capabilities to which we can outsource the computational tasks that classical machines find most difficult. Because the software was designed with this in mind it is possible now to develop outsourced processing hardware that can perform these calculations in near instant time, allowing the classic system to do what it does best - “move stuff”!

That's cool .. so how far off is a practical hybrid quantum computer?

The aim is for 18 months. We have a 6 bit ambient temperature design on the table. That has to be tempered with the fact that Toridion's QNN runs on a 128 tbit (tbit is our secret sauce) simulation, and so we have a way to go yet before we get to a 128 bit device.

What will it be like – how will we use it?

Most people don't want to learn quantum computing to use it .. Most coders want turnkey solutions to client problems in terms of AI and machine learning. To that end we want to deliver our customers a plug and play device, not unlike a USB stick or any other peripheral. Plug it in and run your software. We have already started preparation of our Quantum Cloud API which will allow customers to develop software that uses quantum machine learning via our cloud attached TQNN appliances right away, eventually migrating to on premises solutions that will seamlessly migrate from cloud to desktop.

Once you put this kind of power into the hands of everyday coders you have a whole new paradigm of machine learning capabilities available. Complex convolutional tasks are accelerated by 1000's of times, and the need for such huge volumes of training data (often taking weeks to process) can sometimes be reduced to minutes and hours due to the ability of the QNN to self generate and retrain. In addition to the speed increase you have the energy saving and storage space reduction to factor in. QNN's don't grow with the number of training examples like classic networks, they adapt and sometimes get smaller as they learn...

The future of Quantum Neural Networks is bright, and Toridion is well placed to deliver the next generation of machine learning based on hybrid quantum computing architecture.   


Website powered by Firecart X eCommerce