It is the year 1965, the Ranger 8 probe completes its mission to photograph locations for the landing of the Apollo program and then, it crashes against the Moon. In Spain comes into force the new Press and Printing Law, promoted by Manuel Fraga. And in the United Kingdom, advertising of cigarettes on television is prohibited.
That same year, a young engineer called Gordon Moore, I worked as a director in the laboratories Fairchild Semiconductor. Then he observed a trend in microelectronics that would come to be known as the law that would carry his name: Moore's Law. Later, in 1968, he would create the company Intel together with his partner Robert Noyce.
Although it is not a law in the scientific sense, but rather an observation, it laid the foundations of technological growth – and its cheapening. It is applicable to personal computers and mobile phones, although when it was formulated, microprocessors did not exist and would not do so until 1971. This law stated that each year the number of transistors in a microprocessor It would double. As a result of this rapid evolution, prices would fall at the same time as benefits would increase.
Ten years later, in 1975, Moore slowed down to two years, modifying his own law. And in 2007 it determined an expiration date, stating that in 10 or 15 years a new technology would end up replacing the current one.
In view of this problem, and given the need to process increasingly complex mass data, in Fujitsu, like other large technology companies like IBM, Microsoft or Google, started working on the issue of quantum computing.
"Conventional computing can only do one thing at a time with a group of data. And in quantum computing, thanks to the so-called superposition principle, information can have multiple states at the same time. " David Snelling, director of the Artificial Intelligence program in Fujitsu, explains the benefits that this technology would bring to society and the digital business transformation, since Moore's law is reaching its limit to improve the performance of computers.
But he also talks about the difficulties he must assume: "The challenges facing quantum computing are related to the development of a technology that manages to make it work. It requires an extremely low temperature, only a few milli Kelvin below zero. "He adds:" This computer of the future is very expensive, and at the moment it is only able to solve small problems. This is the true situation with which we are in quantum computing. "
To solve this technological wall, Fujitsu has developed what they call Digital Annealer, something halfway between the present and the future of computing:
"In Fujitsu we have taken a different direction. We decided to take some of the quantum properties and see if we could imitate them, such as superposition or entanglement. "
Unlike the quantum computers, this architecture does not need to operate at an extremely low temperature, but works perfectly at room temperature. That's where the curious name comes from – Digital Annealer – whose literal translation would be "digital annealing", meaning annealing not like an overheated pot, but like the heat treatment used to treat metals. Annealing consists of heating the metal to a certain temperature and then allowing it to cool slowly, until it reaches room temperature. In this way the deformations in the metals are obtained to increase the plasticity, the ductility and the tenacity of the material.
The algorithm mimics these properties with the following example:
We have a lot of pieces of different shapes (Tetris type) that we want to put in a box. With the traditional method of computing, we would need to check the different fitting possibilities by placing the pieces one by one. Starting again every time that combination does not turn out to be correct.
On the other hand, in the "tempered simulation", it would be as if we were shaking all the pieces with force, to gradually diminish this movement at the same time as finding the shapes that fit, much more quickly.
"We have seen that it worked between 300 and 1000 times faster than with a conventional system. So it also achieves 300 to 1000 times more performance than conventional solutions. "Snelling concludes:" We are very excited to know where it will take us. "
However, the director of the Fujitsu Artificial Intelligence program is more cautious about this field: "Digital Annealer is not an artificial intelligence accelerator. You can not take an AI program and put it in Digital Annealer. " However, his contribution does consider it important: "It can be used as part of the segmentation of AI to solve very challenging problems. "