Fujitsu
Not logged in » Login
X

Please login

Please log in with your Fujitsu Partner Account.

Login


» Forgot password

Register now

If you do not have a Fujitsu Partner Account, please register for a new account.

» Register now
Feb 28 2019

The Power of Quantum-Inspired Computing: Journey of Digital Annealer (Part 3)

/data/www/ctec-live/application/public/media/images/blogimages/digital-annealer/Digital_Annealer_Slide.jpg

For most of us, the idea of quantum computing is akin to an exotic beast: Somewhere in the back of our minds, we know that it exists and that at some point in the future we might even be able to build proper machinery, which will be easily accessible and can solve real-world problems that are unsolvable today. But what would you say if someone told you that it is possible to provide and utilize quantum-like capabilities and solve these problems in the here and now? Well, that would seem like this person had just claimed to have spotted Schrödinger's cat in the wild. This five-part blog will help us find out more.

In the first two parts, we introduced a few core principles of quantum and quantum-inspired computing as well as the obstacles that hinder the development of operational hardware. This third chapter describes Digital Annealer, a new platform technology created by Fujitsu to solve combinatorial optimization problems that have so far appeared insuperable, and the underlying architecture.

Before we start, let's do a quick recap of what we covered in part 2 of our mini-series. We spoke about the challenges that researchers into quantum computing found themselves to be up against and how the problems being researched today could be solved from a real-world perspective with Fujitsu's Quantum-Inspired computing.

  • Stability: Existing 'true' quantum computers use qubits to encode and process information. But qubits are unstable by their very nature, so they do not retain their quantum state long enough to carry out any meaningful computation at scale with the existing technology.
  • Accuracy: The lack of stability induces massive accuracy problems, leading to the need for error correction. This is possible to a certain degree, but basically takes away the speed advantage.
  • Infrastructure complexity and affordability: True quantum computers need to be shielded against electromagnetic interference and must operate at a temperature close to 0 °K. Therefore they have to be kept in a unit called dilution refrigerator that is separate from the rest of the data center. Such a system is hugely complex and hence unaffordable for a vast majority of users.
  • Readiness: The quantum systems we have seen so far are fit to tackle sample size problems for research purposes, but do not yet have the capabilities, or scale, to solve real-world problems.

This left us with the following question: Is it possible at all to benefit from the general insights into quantum phenomena and use them to solve some critical real-world problems today?

While classical computers based on digital circuitry are more mature and advanced with respect to usability and energy efficiency, the fundamental, i.e. sequential way the system computes makes it impossible to arrive at the optimal solutions for combinatorial optimization problems. As a result, there is heavy dependence on approximations, which leads to sub-optimal solutions. Leveraging the benefits of both technologies, researchers at Fujitsu have devised a new type of accelerating, quantum-inspired architecture called the Digital Annealer Unit or DAU as the key component of the Digital Annealer Solution.

Inspired by Quantum – The Digital Annealer Solution
In practical terms, the architecture incorporates an optimization circuit, based on digital circuitry, as a building block. Multiple building blocks are driven, in parallel, in a hierarchical structure. This structure minimizes the volume of data that is moved between basic optimization circuits, making it possible to implement them in parallel at high densities using conventional semiconductor technology. In addition, thanks to a fully connected structure that allows signals to move freely within and between basic optimization circuits, the architecture is able to handle even the most complex combinatorial optimization problems.

As its name suggests, the Digital Annealer solution leverages the concept of annealing, the combined heating-and-cooling treatment used in metal processing to first alter the physical characteristics of a given material and then stabilize them in the new state, in the way that an armorer forges a sword. The DAU follows a similar logic of progressive approximation to an ideal state, but instead of craftsmen's experience it takes its inspiration from quantum phenomena. This, in turn, provides massive benefits with regard to the speed, efficiency, robustness, ambient conditions, and cost effectiveness of Quantum-Inspired computing (as opposed to the slowly evolving 'true' quantum computing):

Image

Inspired by superposition, DA evaluates all subsequent candidate states arising from 8,192 possible bit flips. All of these operations are carried out in parallel, which allows for massive speed gains when compared with regular digital processors. These speed gains are similar to what you would get if conventional processors jumped 14 performance generations ahead as per Moore's law.

 

Image

Inspired by quantum entanglement, the DAU provides full connectivity of all 8,192 bits at up to 64-bit precision. Not only does this facilitate near-instant interaction across the entire system in a given state, but it also provides the ability to represent large-scale, highly complex problems effectively and accurately. This allows for easy problem mapping, which again helps to accelerate processing speeds while ensuring the DAU offers higher cost effectiveness than can be found in quantum-annealing (QA) systems.

Image

 

Inspired by quantum tunneling, the unique DAU architecture simultaneously evaluates multiple energy potentials, akin to quantum tunneling in QA, accelerating convergence. Once again, this enables speed gains and cost advantages. Plus, the DAU increases the escape probability from a locally optimal solution.

 

This unique digital architecture inspired by the above mentioned Quantum phenomena provides compute speed and accuracy in a conventional datacenter environment and also the following benefits:

  • Digital Annealer is able to operate at normal room temperatures and is easy to miniaturize, so as to fit into data center environments and doesn't need a separate dilution refrigerator. As a consequence, it provides much higher energy efficiency and incurs much lower energy costs than true quantum computers.
  • Digital Annealer provides 8192-bit full connectivity, allowing all bits to freely exchange signals and enabling the platform to deal with real-world, large-scale problems.
  • The DA solution supports a common tooling platform to that of quantum-annealing systems, making it easy for existing customers to qualify for quantum computing when this technology matures.

We should also add that we are using quantum algorithms to perform Digital Annealing, a computing procedure that forgoes regular programming steps and instead relies on setting a few parameters to trigger calculations. To the uninitiated eye, this integrated end to end solution operates automatically and almost completely independent from human interaction save for some basic input, i.e. a selected collection of data and the parameters just mentioned. The DA can be consumed remotely or deployed on-premises, using REST API interfaces and SDK's.

In part 4 of our mini-series, we will see how DA solution enables organizations to make breakthrough changes by solving large-scale, time-critical optimization problems to not just massively optimize their business, but also to create a new disruptive market in their field.

 

ImageImage

Manju Annie Oommen

Dr. David Frith Snelling

 

About the Author:

Manju Annie Oommen

Sr. Manager – Product Marketing

About the second Author:

Dr. David Frith Snelling

Fujitsu Fellow and Program Director Artificial Intelligence, CTO Office, Fujitsu

SHARE

Comments on this article

No comments yet.

Please Login to leave a comment.