As the business world continues to grow exponentially, the leaders who build, create, and offer solutions also are matching the pace by establishing next-level experiences. Be it market research, product development, marketing, trends, analytics, and more, promising leaders are honing their business and life skills to innovate and disrupt the space.
One such relentless leader who is leading by example is Axel Kloth. He is a physicist and computer scientist by training and the Founder, President, and CEO of Abacus Semiconductor Corporation. He was also the Founder and CTO of an AI-enhanced security processor company, for which he developed the underlying technology and was instrumental in the fundraising of around $20M in two rounds.
Unfolding the Journey
Axel Kloth studied Physics with Information Technology (“Informatik” in German). As a leader and entrepreneur, he believes a good and solid education lays excellent groundwork for any scientific and engineering work. Without it, everything becomes a bit more difficult.
While he has not used his Physics background to uncover new physical phenomena, the underlying understanding and the math needed help him optimize its products to solve customers’ problems. As he uses supercomputers, he knows what the user’s level of frustration can be. Hence, he utilizes that along with his customer’s inputs to build better processors, better accelerators, and better shared multi-homed memory subsystems to build better supercomputers.
Axel Kloth is a serial entrepreneur with an extensive knowledge of technology and its business potential. Before founding the AI-enhanced security processor company as a spinout, he started SSRLabs (Scalable Systems Research Labs) with a focus on HPC, and Parimics, a vision processor and systems company that in 2004 built the foundation of real-time object detection, motion analysis, and object tracking in hardware, with as many objects trackable as there were in the frame set.
Parimics, in 2004, was the first company to use a convolutional neural network processor for object detection, and a second processor with matrix, tensor, and transform math acceleration for linear and non-linear object tracking and path prediction. All of those technologies are today considered the gold standard for AI and ML. Axel holds the patents “Method and apparatus for image processing”, Patent number 7564996, and “Method and apparatus for image processing”, Patent number 7489834.
Axel Kloth was involved in several startups in optical communication, Distributed Feedback Dye and semiconductor LASERs, high and continuous availability systems, and resilience and survivability studies for exposed systems as well as in projects for the US Department of State. Additionally, he is a Partner at and serves as an advisor to Pegasus Tech Ventures, a global VC firm that conducts the Startup World Cup Challenge.
Axel Kloth has been involved in cryptographic research (before NIST created the crypto challenge and subsequently defined and published crypto standards) and securing data in transit and data at rest for nearly as long as he has been using and developing solutions for HPC. He is known as a disruptor and a creative mind to come up with new solutions, and he particularly likes to prove people wrong if they claim something is undoable.
To that effect, he was behind the world’s first CML high-speed serial links on bulk CMOS, which was believed to be impossible to do. He is also behind the world’s first combined virtually output queued (CVOQ) switch fabric and many other inventions in the areas of processor architecture, communication I/O, protocols, cache coherency, and I/O offload.
Axel Kloth’s current focus is HPC. Many experts in the HPC community are raising flags about the current direction of the industry as Moore’s Law came to an end, General-Purpose CPUs don’t show drastic improvements in absolute performance and performance per Watt anymore, and the accelerator space is very fragmented. More importantly, there is no universal interconnect between processors and accelerators commensurate with their performance levels. Axel identified this trend a long time ago and is working on solving this issue with better (more linear) performance scaling that will lead to better HPC.
A Take on Business Leadership
Axel Kloth reckons that Business Leadership has changed drastically in the past few years. While many areas continue to thrive, certain areas experienced a lack of available talent, and as such it has become mandatory to build teams and retain them. With more and more knowledge workers being the backbone of High-Tech companies, it is essential that knowledge is retained within a company, and that this knowledge becomes institutional.
If a substantial part of the technology development team leaves, then the company is going to face not only a shortage of workers that can execute a task, it will experience a brain drain that is hard to undo. Consequently, building a team and retaining it is more important than ever. This can only be achieved if the team feels validated and acknowledged.
Confronting and Conquering Challenges
Challenges are a part and parcel of every journey. As a founder, one has to analyze the obstacles, find optimal solutions, and integrate them according to the need of the hour. Every day brings in a new challenge but with a robust mindset and adept nature, one can successfully conquer it.
According to Axel Kloth, challenges always include convincing others, particularly if that subject matter is not scientifically accessible, and if engineering disciplines are not applied very rigorously. Another challenge is always the funding side of the business, and fabless semiconductor companies are notorious for a very high barrier to entry for newcomers, and large upfront investments. However, once a solution is in the market, these barriers become a huge positive as it keeps competition out.
Stepping into the Shoes of a CEO and Founder
As the Founder and CEO, Axel Kloth handles many responsibilities. He is thoroughly responsible for the well-being of the company and its team members. This includes raising funds and building the teams that do the work, and of course, hiring the respective managers to supervise the teams. This has become a vastly more complicated task as stated before—in High-Tech companies, team building and retention is crucial, and many old approaches don’t work anymore. As a result, hiring the right supervisors is crucial.
Bestowing a Culture of Encouragement and Empowerment
A team is an integral part of an organization. And being a team player, Axel Kloth recognizes the importance and unity of a team for holistic organizational growth and success.
He states, “We empower everyone. We ask them to solve a problem and never get in the way of how they work. The results count, not how they got there. If they want to share how their outcome and the way to get there is novel and better, we encourage that. We also try to never provide solutions if an employee gets stuck on a task – instead, I try to find out where and why they are stuck and then ask questions so that they can get themselves unstuck. That seems to work a whole lot better than giving them an answer.”
Cherishing Moments of Victory
Among others, Axel Kloth wrote a few books and authored or co-authored about 40 patents, developed the first and only working Combined Virtually Output Queued (CVOQ) switch fabric, and built the first CML-type transceiver for High-Speed Serial Links (HSSLs) on bulk CMOS which is the underlying technology today for XAUI and PCIe and many others. He also conceived of a new way to guarantee authenticated secure startup (aka “boot”) processes for computers.
He has built ccNUMA (cache coherent Non-Uniform Memory Access) switch fabrics for servers with better path setup procedures and reduced the metadata (cache coherency) traffic and many other technologies related to supercomputers.
Axel Kloths Source of Inspiration
Axel looks up to people who had novel ideas in the past and created products and companies based on these novel ideas. In his case, he truly admires Stan Mazor – one of the four co-inventors of the microprocessor – because his journey was not a one-off.
He went on to spawn the Electronic Design Automation industry because his insight into the microprocessor and the way it was designed made it clear to him that this must be done with the help of specialized software for all future generations of processors and ASICs (Application Specific Integrated Circuits).
About Abacus Semiconductor Corporation
Abacus Semiconductor Corporation was formally incorporated in September 2020, but it builds on decades of technology development that Axel Kloth has been involved with. The depth of technology and IP that Abacus has is greater than what would be expected of a 3-year-old company.
The vision was simply to build better components such as processors, accelerators, and smart shared multi-homed memory subsystems that would allow a system integrator such as Cray to build a better supercomputer. After all, living in the age in which ChatGPT will change the world and its underlying models will continue to increase in size. For GPT-4, the assumption is that 1 trillion input parameters are needed. Processing these huge numbers of parameters takes time and a large degree of parallelism that is efficient enough to reduce the compute time to reasonable periods.
No one will want to wait for a new GPT model for 12 months. Any changes to the input parameters should be able to be incorporated within days, not months. That requires new thinking – which is what Abacus does. With better supercomputers, the world can build digital twins of any system that society relies on, and then test out what inputs are needed for optimal outputs. The company hopes that its products are used to build those supercomputers so that humankind can make much better decisions in the future, to enable not only survival but thriving humankind.
An Overview of the Products Offerings
Intel, AMD, and ARM as well as NVIDIA are building fantastic products for applications that rely on crunching what are called scalars and vectors, and for most general-purpose applications. However, the computational requirements for Artificial Intelligence, Machine Learning, and particularly GPT (Generative Pre-training Transformer) converge with traditional HPC in the sense that it needs more performance for matrix and tensor math on large-scale input data. This is what Abacus does best. Its processors, accelerators, and smart shared multi-homed memory subsystems focus on exactly that.
Abacus focuses on what it does best – large-scale computers for large-scale problems. Axel Kloth adds, “If you use cloud services today and are happy with the results, you likely won’t need our products. However, if your success is predicated on using ChatGPT with continuously updated GPT models, then you will need to use our technology to build the model creation and updates. Similarly, if you have a very large computational problem that needs to be solved in a reasonable time frame, and current solutions even in the cloud don’t help, then you will need our technology.”
A Word on Change, Work-Life Balance, and Advice for Business Aspirants
Change is the need of the hour and one has to keep themselves updated with the evolving market dynamics. Axel Kloth does it by reading what others do, trying out what sticks, and if it fails, quickly reverting to novel ideas then try those out, and if they render better results, keep them.
As a Founder and CEO, Axel Kloth states that there is no work-life balance, and if one considers starting a company, take that into account. Additionally, Axel shares a piece of advice for young business leaders to make it big in their careers. He says, “Do what you are good at and keep improving. Stagnation will kill your career, so make sure to continue to learn, improve and understand.”
Optimizing the Future with Best-in-Class Products
Abacus will continue to develop its processors, accelerators, and smart shared multi-homed memory subsystems to cover more computational challenges that need more performance for matrix and tensor math on even larger input data than what was needed for GPT-3 (175 billion input parameters) and GPT-4 (1 trillion input parameters).
Axel Kloth concludes, “We will build on what we have and will create an ecosystem around us that can tie into our infrastructure, and we will do that by augmenting our products with solutions from other companies that cover solutions to computational challenges that we do not offer.”
