Fujitsu Solutions Lab is Breaking Silos, Positioning for an AI Future

By on
Fujitsu Solutions Lab
Fujitsu Solutions Lab

Since breaking away from the Fuji Electric Company in 1935, Fujitsu has established itself as not only a household name but also as an information and communications juggernaut synonymous with technological innovation. This has been evidenced in its multi-faceted ascent through the decades, which saw the company grow to 132,000 employees worldwide, reaching just shy of four billion yen in revenue, while attempting - and succeeding - in many areas, including computer and semiconductor development, home electronics and electromechanical components.

Now? Fujitsu is poised to take on artificial intelligence (AI) and other emerging technologies with their Solutions Lab, a project which aims to break down silos and foster collaboration around technologies including multi-cloud, datacenter scale and efficiency, 5G and, of course, AI.

This isn’t the first time Fujitsu has taken on such a huge technological challenge, with the company developing the K Supercomputer in 2011. The K supercomputer’s technology, based on a distributed memory architecture with over 80,000 compute nodes, was ranked as the world’s fastest supercomputer by the TOP 500, an industry authority that aims to improve and renew the Mannheim supercomputer statistics.

The K Supercomputer’s 10+ petaflop - or 10 million billion calculations per second - computing speed has proven useful in a variety of crucial applications, including climate research, disaster prevention, and medical research. However, In June 2012, the K Supercomputer was superseded by the American IBM Sequoia.

Since then, Fujitsu has continued to develop stronger and more accessible technologies, like the currently under-development Post-K supercomputer, which aims to be the most advanced general-purpose supercomputer in the world. Through inheriting and advancing the K Supercomputer’s innovative features, Post-K aims to improve application performance, reduce power consumption, and enhance user convenience while also offering the ability to produce groundbreaking results.

The Post-K infrastructure also aims to encompass brand new applications, including numerical software and deep learning capabilities.

“Current trends and evidence suggest deep learning and artificial intelligence is the way of the future,” says Alex Lam, Vice President and Head of the North America Strategy Office at the Fujitsu Solutions Lab. “As such, our Solutions Lab has conceptualized new architecture to facilitate deep learning, backed by the Post-K infrastructure, called the Deep Learning Unit.”

The Fujitsu Deep Learning Unit, also known as DLU, is chip architecture purpose-built to facilitate deep learning. This under-development architecture aims to upgrade conventional AI architecture’s complicated general-use cores to domain-specific counterparts; deliver optimal precision by upgrading double and single precision floating points to a deep learning integer, and is aiming to be massively parallel, upgrading from multiple sequential and parallel cores to many, massively parallel cores within an on-chip network.

NVIDIA has been in the headlines with its stock price rise due to demand for graphics processing units (GPU) for enterprise purposes, previously used for video game performance. But these GPUs have not been the optimal answer to the demands of today’s computing needs. Thus, Fujitsu developers began work on a new architecture which resulted in the DLU confronting a key challenge: the ability to deliver high-performance in deep learning with the reliability of mainframes while lowering power consumption. Lam stated, “a challenge in developing any new technology is the balance between performance and power consumption. The traditional logic states that the higher a computer is performing, the more resources - or power - it needs to perform at that level. Reworking this model is no easy feat.”

“To address this, Fujitsu has reworked the old architecture integrating heterogeneous cores, distributed processing environment (DPE) and large register file (RF) chips, with Fujitsu's deep learning integer to realize an efficient deep learning engine that blends the reliability of a mainframe with 50 percent less power consumption,” Lam adds.

While direct applications for these AI capabilities have not been stated by Fujitsu developers, it’s widely understood that the applications of supercomputer-backed AI technology are seemingly limitless - from advancing knowledge in areas like climate modeling, to designing new aircraft and creating new materials.

Fujitsu’s innovation aims don’t stop at deep learning and supercomputers, with the Solutions Lab also pioneering research and development in 5G, edge computing, IoT, and data and analytics. The Silicon Valley-based lab has the resources set up for proof of concept (PoC) testing to enable technology partners and end-customers to integrate their technologies with Fujitsu solutions in an effort to achieve digital transformation and overcome next-generation data challenges.

Lam believes the pathway to achieving this digital transformation is through a four-stage process, stating ”The first is by showcasing Fujitsu IT and technological innovations to attract technology partners who wish to transfer their knowledge and integrate our technology into their solutions. The second is by building a technological knowledge-sharing ecosystem with leading next-generation data center industry partners, and the third is to collaborate and develop new technology ecosystem solutions for the next IT age.”

And the fourth? Lam believes their PoC testing infrastructure is what sets the Fujitsu Solutions Lab apart from other, similarly angled technology incubators, with the lab designed for not only in-person PoC testing, but equipped for remote testing, stating, “Instead of having to ship equipment to a geographical location, Fujitsu partners are able to conduct PoC testing remotely. We have been met by some cynicism around this, but with our infrastructure, developers can run a real workload without having to be at the actual Solutions Lab.” Enabling developers to access the Solutions Lab PoC testing infrastructure remotely is likely to increase accessibility while reducing the associated travel, shipping and time costs.

Lam believes this accessibility, combined with an advanced technological infrastructure, knowledge, and the forward-thinking mindset of developers and partners have Fujitsu poised to not only take on artificial intelligence but to help shape the world’s future.

Join the Discussion