As the world’s reliance on technology is increasing, the need for supercomputers is rising in every sector. Organizations need supercomputers that are future-ready, cost-effective, energy-efficient, and time-saving. However, the existing supercomputers often take at least a week to complete a complex problem when a one-hour response time is needed.
Addressing this more-than-order of magnitude gap between the theoretical peak performance of supercomputers and the real-life performance that users of today’s supercomputers experience is a revolutionary semiconductor company—Abacus Semiconductor Corporation (ASC).
A vast majority of today’s supercomputers are all built based on commercial off-the-shelf (COTS) servers with industry-standard x86-64 processors and GPGPU accelerators, without any regard to high-bandwidth and low-latency communication between processors, accelerators, and memory. As a result, their performance does not scale out even remotely linearly over the number of cores – and that at a time during which we need more supercomputing performance than ever. The existing underlying architecture has few limitations and challenges and must be revised.
ASC is solving this problem by rethinking the connections between processors, accelerators, and memory and by providing a novel interface to connect all of these components. An open-source Instruction Set Architecture (ISA) based processor, special-purpose accelerators for Machine Learning and traditional supercomputer applications; and the ability to make memory smarter and faster help ASC accomplish this goal.
Enabling ASC to stay ahead with a focused approach and innovation with his thirty-plus years of experience in the industry is the company’s resourceful Founder, President and CEO—Axel Kloth.
Let’s get to know more about the ASC’s journey and how it is reimagining the semiconductor industry with the standout technology leadership of Axel Kloth.
Solving Tomorrow’s Challenges
Abacus Semiconductor Corporation is a fabless semiconductor company, meaning that, the company designs and engineers its products but lets a world-class manufacturing company produce its products. They then ship these products to ASC’s customers so that they can build their servers and supercomputers with them.
While that is typical for fabless semiconductor companies, in most other aspects, it is not typical. “We develop all of our crucial IP in-house and pride ourselves on not having to buy IP for any critical components. This keeps intellectual property in-house and allows us to build institutional knowledge,” highlights Axel.
This allows the company to optimize the scale-out of the performance of its processors, accelerators, and memory subsystems nearly linearly. Unlike established processor companies, ASC makes the integration of its and other companies’ accelerators easy.
As described by Axel, in traditional supercomputers, it is generally observed that 100,000 servers provide the performance of about 5000 to 7000 servers in most non-benchmark applications, and that efficiency drops even further if accelerators such as GPGPUs or Artificial Intelligence/Machine Learning (AI/ML) engines are included.
In ASC’s case, the company anticipates a scaling factor of over 80%. This gives the user more useful performance for their applications, lowers power consumption, and lowers the cost of running the supercomputer. “This is particularly relevant today as traditional CPUs and accelerators have proven to not work well in current supercomputer architectures, and quantum computers are not commercially available yet for the applications that the industry needs to solve challenges in the next 10 to 15 years,” expresses Axel.
The Guiding Mission
Abacus Semiconductor Corporation’s mission is to broaden the deployment of High-performance Computing (HPC) by making it cheaper, easier to use, and more energy efficient while at the same time improving performance. “We want to enable our customers to build personal supercomputers so that scientists requiring intense computational capabilities to simulate the outcome of decisions affecting the viability of human life can do so in an easy-to-use way,” explains Axel.
Furthermore, Axel mentions that there will be nine billion humans on planet Earth in less than 30 years, or a generation’s worth of time. Feeding those people will require growing and distributing the food with minimal waste and spoilage. He says that organizations worldwide also need to ensure everyone can live a healthy life and get unrestricted access to educational opportunities so that everyone can play a better role in reducing the global birthrate. Simultaneously, Axel opines, “we must reduce CO2 production so that global warming can be stopped, or ideally reversed.”
To do so, Axel believes, “we must be able to simulate the outcome of each decision prior to implementing it, and that requires us to simulate the impact at a faster rate than real-time. This will enable us to make more decisions in furtherance of those goals while avoiding the negative impacts of decisions made with incomplete information.”
Adding up, he says, “While Earth will continue to exist without us, the survival of humans is entirely dependent on our decisions.”
He continues, “The more decisions we make to support human life and sustainability, the more likely it is that our species will survive. Abacus Semiconductor Corporation wants to be a key participant in this process.”
The Robust and Cost-effective Offerings
ASC is working on a Server-on-a-Chip processor that will provide Internet backend functionality (using a Linux/Apache/MySQL/PHP or LAMP stack) at a lower cost and lower power consumption per executed instructions than competing offerings. This processor can also be used as an I/O frontend for a supercomputer, and it easily connects to more of the same processors through a very high bandwidth and low-latency interface. That same interface connects it to a higher-performance memory subsystem, which ASC is developing, and a set of accelerators for math and database operations.
The company is making this interface available to others through licensing and through a new trend that the industry calls a chiplet. The benefits of the solution are vastly better scale-out and robustness against cyber-attacks. “We are often asked why scale-out is important, and our answer is always the same analogy. In the workplace, we assume that not everyone can do everything,” says Axel.
He puts, “We have specialization in jobs, talents, and assignments. If an engineer is assigned to invoicing, that’s usually not working out well. If the task needs to be finished in three months but the total design effort is 36 working-months, then we’ll need to put 12 engineers on it. This would be a linear scale-out, and we can plan for it. If we need to put not 12 but 24 engineers on the task, we have a non-linear scale-out, and that is not good as it is cost-prohibitive. The same applies to computers.”
“If you need ten times the performance of your computer to solve a large-scale problem, you should only need to buy ten computers to solve the problem in the allocated time. If you need to buy 40 (or even 100), then your problem was such that it did not allow for a linear scale-out of performance,” states Axel.
Adhering to the Values
Having a focused approach is a must for every organization to keep on progressing no matter what. Having a strong set of core values helps a company to stay centered on its mission. Speaking about the foundational values of the ASC, Axel states. “We strive to live fully by our values. As imperfect beings, we may fall short occasionally – but we pledge to accept our failures, learn from our mistakes, and continually improve.”
ASC does not discriminate on the basis of race, religion, color, sex, age, national origin, physical or mental characteristics or disabilities, sexual orientation, gender, or gender identity, marital, parental, or veteran status. ASC values diversity in all of its facets. The company is committed to inclusion at the highest level of management and the development of leadership from historically underrepresented groups (e.g., women, LGBTQ, people of color).
Moreover, the company has pledged to actively listen to stakeholders’ perspectives and criticisms. In fact, it encourages lively, fact-based arguments. Axel adds, “We value hard work, high ethical standards, strong morals, and active engagement in environmental causes.” The company has pledged to hire and promote based on merit, using objective and subjective criteria in a transparent review process.
Illustrating the values further, Axel tells us, “We act on facts. We operate based on data. We speak with numbers and are intolerant of misinformation.” He further says, “We firmly believe in democracy and its promise of self-determination. Though democracy may be imperfect, it is the best form of government that humans have invented. We strive to improve on the foundation we were fortunate enough to inherit.”
Strengthening Technology
To deal with modern problems, technological solutions must be efficient and ready for challenges. Sharing his views on the technology requirements of modern times, Axel says, “Large-scale compute, also called High Performance Computing, is needed for a variety of challenges that humankind faces today.”
He feels that “The more easily HPC can be used, the cheaper it is, and the more accessible it is, the more we open up HPC to users who make decisions for companies, for policies, and for humankind in general.”
He adds, “Ultimately, HPC, including AI and ML, can help people make better decisions, whether these people are in government, company leadership positions, or in Research.”
Rising Above the Challenges
Explaining the challenges within the industry, Axel says that semiconductors show somewhat of a resurgence, but the general funding level from VCs and private equity does not quite reflect that yet.
“We are on our path to raising our A round, but it certainly has been more difficult than in the years up to 2021. At the same time, solutions are needed urgently. Traditional HPC solutions do not work anymore with the necessary inclusion of accelerators such as GPGPUs and special-purpose engines for AI and ML, but at the same time, quantum computers are not yet relevant and they cannot solve current computational problems or the ones we expect in the next 10 to 15 years yet,” he expresses.
“That is where we come in because we believe that we can solve those problem sets, even after quantum computers start being deployed and used to solve real-world problems.”
Showing the Path to Industry Aspirants
Axel shares valuable industry insights with the aspirants, saying, “Semiconductors are the foundation for most devices today, whether in industrial or personal use. As such, there will be a continuing need for semiconductors in a wide variety of applications, features, sizes, and cost targets.”
He adds, “The need for processors, microcontrollers, and more IP building blocks will continue to grow, and as a result, I believe that there will be a demand for ASIC design engineers for a very long time to come.”
A Look at Emerging Trends
Speaking on the trends that are emerging in the industry, Axel observes, “The semiconductor manufacturing industry will continue to consist of the same players that are out there today, but the new fabs will be built in politically less contested areas.”
Furthermore, he predicts, “Specialty and analog/mixed signals have not yet been consolidated, but that might happen soon as well. Semiconductor design will be re-homed to the US and Western Europe in the long term.”
The Primary Goals
Axel describes ASC’s roadmap to further its growth in the coming times and quotes, “We anticipate that we will continue our growth. Our product roadmap is validated by customers, so now our primary goal has to be on implementation, integration, and on firmware and OS kernel design that align with our customers’ needs.”
The Proficient Leader
Axel is a physicist and computer scientist. He has been using supercomputers for over 30 years, and at the same time, he has been designing components that go into supercomputers for about the same time. That mix of experiences has allowed Axel to identify what’s working and what’s not and then focus on fixing the components and architectures that pose a problem.
Axel’s research has yielded verifiable results in the areas of processor pipelines, scheduling, ALUs, switch fabrics, and high-speed serial links, caches and cache coherency, memory architectures, and multi-processor systems and their interconnect. By combining these, Axel has created an unprecedented, novel system architecture that the established players cannot match.