Abacus Semiconductor Corporation: Venture with a Distinctive and Visionary Approach

10 Innovative Brands of the Year 2022

Below are the highlights of the interview between ASC (Abacus Semiconductor Corporation) and Business Leaders Review

What Abacus Semiconductor Corporation is aiming to achieve?

Abacus Semiconductor Corporation (ASC) targets the scale-out problem in High-Performance Compute (HPC). In today’s supercomputers, thousands of servers are connected to solve a single problem. Most computational problems require a large amount of connectivity between processors, and unfortunately, this is not baked into the design of current processors powering those supercomputers. While processor cores have improved very drastically in performance over the past 20 years, the interconnects across processors have not. To illustrate that, imagine that two employees sit at a shared desk. If one of them needs help, he can ask his colleague, and very likely, the colleague will be able to help out very quickly. Now imagine these same two employees sitting at desks on the fourth floor in different buildings, and the phone does not work. Now every single time that there is an issue with clarifying a task or a question about anything, one of them has to go down four flights of stairs, cross the street, go up four flights of stairs and then ask his colleague.

Obviously, in this scenario, the second set of employees is going to be a whole lot less effective and efficient, though it’s none of their fault. That is how supercomputers today work.

We fix this through a combination of more connectivity, different use, and attachment of memory, and a unified way to connect processors, memory and accelerators, particularly those for Artificial Intelligence, Machine Learning and traditional HPC workloads. In other words, we have designed a new processor, a smarter memory subsystem, and a set of accelerators all predicated on a novel interconnect that is faster and has a much lower delay (or round trip time or latency).

Why the name Abacus?

We chose the name Abacus because the Abacus was one of the first mechanical devices to help humans speed up processing math problems while making it more error-proof.

Shed some light on the different sectors and service platforms ASC caters to.

We anticipate that our technology will first be used in supercomputers that are targeting specialized computational problems that cannot be solved otherwise in a timely fashion. First certainly-premises, but ultimately also in the Cloud as HPC-as-a-Service offerings, and then as a hybrid solution with on-premises and Cloud-based installations combined.

How has technology changed the digital and material world? And also, how does Semiconductor behave as a catalyst?

With the technology we are developing, we believe that researchers can simulate many more computationally hard problems that we need to solve to maintain life for humankind on this planet. Sustaining 10 billion humans on this planet requires careful use of resources, generating food and distributing it to everyone, doing the same for drinking water, providing shelter and giving them access to a roof over their head and good education.

All of this requires planning, simulation and so-called genetic algorithms to optimize the approach in a computer first before implementing it in the physical world. At the same time, we need to prepare for the next pandemic and have the computational performance available to identify the weaknesses of the next pathogen that we then can exploit to eradicate it. Engineering better batteries, better anode and cathode materials for them as well as a better understanding of how and why the batteries degrade over time to avoid shortened life spans creating more e-waste is another important computational problem that can be solved with more supercomputers in the hands of more researchers.

Elucidate on the early stage of ASC and motivation to drive the initiative.

The Founder started ASC simply because he is a user of HPC and has designed for HPC. However, he is still frustrated by how demanding and expensive HPC is. ASC is set to solve these problems by making the scale-out much more linear and avoiding the problems as indicated above.

What challenges and lessons have ASC learned while catering to the sector?

Along the way, we have learned many things that our customers have given us feedback on. Most often, we hear that our customers have the same problems that we have encountered, and they agree that the problem highlighted exists for them as well, and most importantly, they agree that the solution we propose appears to solve the problem. We have not heard any pushback on the technology. We often hear that our processors are not compatible with the mainstream Instruction Set Architecture (ISA) that is currently the prevailing instruction set used in most supercomputers.

 It is true that the largest number of supercomputers is based on what is called x86-64 (the 64-bit version of the x86 processors from Intel and AMD); however, ISA was not always dominant. In fact, the fastest supercomputer in the top500.org list has been based on nearly every ISA that the industry has created over time, and today’s fastest machines are all predicated on accelerators that use a unifying programming interface called CUDA. That simply means that most of the computing is not done in the processor, but in a set of accelerators, making the ISA irrelevant. At the same time, it means that commercial off-the-shelf (COTS) servers can and will be based on any processor that solves the given problem at the lowest cost and with the lowest power consumption for the performance in the targeted application, at the level of scalability needed to solve the computational problem.

How do you quantify and qualify your team? How is it benefiting the facility services?

Naturally, to develop a system that is drastically different from what others do to solve the scale-out problem, we focused on hiring the brightest and the best, without regard for the pedigree of the schools they visited. This is why all of our contributors are non-conformists. We needed to break the old way of thinking, and that is not possible with people who are stuck in a rut. As a result, our contributors all the way up to the Technology Advisory Board, the Strategic Advisory Board, the Board of Directors, and the Executive Management team are as diverse as it gets, and we are proud of it.

Briefly describe the lead (person) of ASC who takes care of most of the work.

As the Founder and CEO, it is up to me to distribute tasks to individuals and to groups so that everyone can contribute to the best of their abilities to finish the development of our processors, accelerators and smart memory subsystems while building the company. Designing and engineering a set of products that are different from mainstream products requires a team effort. This cannot be done by an individual alone; it must be finished as a team effort.

How did you cope with the pandemic? In what way did the pandemic hit you?

The pandemic hit us the same way it hit other companies and teams – collaboration in an office setting with multiple engineers on a table became impossible. Remote work to some degree helped us overcome some, but not all challenges associated with it.

What is the future of the Semiconductor based industry and how are you preparing yourself to cope with the revolution?

We believe that our technology will span the gap between now when supercomputers, as built today, will not be able to cope with the ever-increasing demand for computational performance, and the time at which general-purpose quantum computers will be able to take over. We envision working with special-purpose quantum computers as accelerators for a while as there is no consensus in the industry about when general-purpose quantum computers will be technically and economically feasible. Until then, we believe that we have the best solution