How Ceremorphic is disrupting the semiconductor space with its patented technology and new reliability architecture

 Ceremorphic was founded in April 2020 by Industry-Veteran Dr. Venkat Mattela, the Founding CEO of Redpine Signals, which sold its wireless assets to Silicon Labs, Inc. in March 2020 for $308 million. Under his leadership, the team at Redpine Signals delivered breakthrough innovations and industry-first products that led to the development of an ultra-low power wireless solution that outperformed products from industry giants in the wireless space by as much as 26 times on energy consumption.

The rapid proliferation of machine learning applications and their inherent high-performance demands, have made reliability a mainstream issue in semiconductors. Today, as we all know, a specific set of applications like automotive have adopted reliable silicon in their system designs. Ceremorphic is designing an ultra-lower power supercomputing chip built-in TSMC 5nm, leveraging its own patented technology and multi-thread processing architecture, ThreadArch®. Ceremorphic’s Energy Efficient AI Supercomputing has built a new architecture that delivers the performance required for next-generation applications such as AI model training, HPC, drug discovery, and metaverse processing.

With Ceremorphic’s Reliable Performance Technology, companies can train their models in a deterministic and cost-effective way. The ground-up architecture features low power consumption and industry-best-reliability technology to ensure uninterrupted application execution and along with it various safety certified silicon products enable the building of multiple platforms for industrial and consumer markets. Dr. Venkat Mattela, the Founder & CEO of Ceremorphic in discussion with CXOToday share more insights on the same.

• Ceremorphic’ s journey and inspiration

Ceremorphic was established in April 2020. In March 2020, Redpine Signals, a start-up that created an ultra-low-power wireless solution that surpassed solutions from wireless industry giants in energy efficiency by up to 26 times, was sold to Silicon Laboratories, Inc., a US semiconductor business, for $308 million. For more than 15 years, Redpine has worked on advanced wireless technologies, producing market-leading System-on-Chip (SoC) devices that are now being used  in leading smart watch  products. We developed over 130 core wireless patents at Redpine Signals, Inc., which we sold to SLAB for $308 million. With Redpine’s success in terms of innovation and surpassing previously thought-to-be-impossible milestones, we decided to do the same in the AI training supercomputing arena, culminating in the birth of Ceremorphic.

In response, Ceremorphic has already created several ground-breaking breakthroughs. Ceremorphic has identified the most critical necessity in today’s high-performance computing era: reliability and energy consumption. Although Ceremorphic was formally established in April 2020, R&D activities started three years before, and the business currently has over 100 patents in its portfolio. With 150 workers and a highly experienced and trained technical staff, Ceremorphic is a mature AI training supercomputing technology firm. It possesses patented technologies, extensive technical know-how, and an experienced  management team. The business contends that a sustainable product is only conceivable if it develops all the essential technologies internally and owns and distinguishes the underlying technology.

• How is Ceremorphic’s product likely to disrupt the semiconductor space? Is the Indian ecosystem ready for semiconductor manufacturing? Perspective.

The semiconductor technology roadmap comprises tools and nodes for building advanced silicon devices, with the 5nm node being especially efficient for producing high gate density (about 40M gates in 1 square mm area) and excellent performance. With access to this sort of technological node, a developer  may construct incredibly efficient and high-performance CPUs in a reasonable physical size. However, managing leakage power and reliability is difficult with these nodes. These two concerns, reliability and energy efficiency, are currently severely restricting the adoption of AI / ML applications across all industries.

Ceremorphic’s current technology portfolio, as well as the design of QS1 (the first chip in the Hierarchical Learning Processor™  – HLP series), will address reliability and energy efficiency concerns, laying the foundation for an architecture capable of Exascale performance while retaining a manageable power budget. This technology will lower the cost of high-performance computing systems, making AI and machine learning applications more broadly available. When it comes to manufacturing in India, two elements must be considered: skill and infrastructure. Talent is never in short supply in India, but infrastructure and prior demonstrated expertise are required to achieve any significant accomplishment at scale. Both features of the ecosystem are improving noticeably in India.

• Why is energy efficient supercomputing becoming popular? Why is the initiative to improve energy efficiency in HPC/supercomputing important?

Energy efficiency is becoming increasingly important in supercomputers as machines get more powerful and demand more electricity. High-performance computers’ (HPCs) capacity to solve complex workloads quickly has increased greatly since their inception in the 1960s; unfortunately, so has their power consumption. Many supercomputers require more than a megawatt of power, and annual electricity costs can easily exceed millions of dollars. With the unprecedented demand and growth of AI / ML applications and their underlying training computing needs call for a computing paradigm which is not only energy efficient but also be reliable as the number of processing elements in these machines are many fold compared to what we have seen in the past.  Ceremorphic saw a need to improve the energy efficiency of supercomputers and the infrastructures that support them as the use of HPCs rose. Because supercomputers are frequently used to run AI programs, the term “supercomputing” has come to mean “AI.” This is because AI/ ML  training  requires high-performance processing, which supercomputers provide.

Ceremorphic has created a novel architecture for next-generation AI model training, high-performance computing, drug discovery, and metaverse processing. Our ground-breaking silicon geometry-based architecture meets high-performance computing requirements for reliability, security, and power consumption at scale. Ceremorphic has a patent portfolio and internal R&D capabilities, and it is developing a new and scalable computing paradigm that will propel the industry forward for several years.

• Kindly highlight the technology behind delivering Reliable Performance Computing in an Energy-Efficient AI Supercomputing Chip.

Ceremorphic employs its own unique ThreadArch® multi-thread processor technology in conjunction with cutting-edge new silicon algorithm technologies. The team has developed an ultra-low-power training supercomputing processor leveraging its comprehensive knowledge and unique technologies. The issues that this industry faces with ‘reliable performance computing’ cannot be met with existing designs; instead, a whole new architecture designed expressly to deliver reliability, security, energy efficiency, and scalability is required. Ceremorphic’s approach is a huge step in the right direction for this business, which sorely requires dependable performance computing.

The Hierarchical Learning Processor (HLP) selects the best processing system at a given time in the execution for optimal  power performance metric . The following are some of the QS 1’s key features:

• Custom Machine Learning Processor (MLP) running at 2GHz
• Custom FPU running at 2GHz
• Patented Multi-thread processing macro-architecture, ThreadArch® based RISC –V® processor for proxy processing (2GHz)
• Custom video engines for Metaverse Processing (1GHz) along with M55 v1 core from ARM®
• Custom designed X16 PCIe 6. 0 / CXL 3.0 connectivity interface
• Open AI framework software support with optimized compiler and application libraries
• Soft error rate: (1000,000)-1

Ceremorphic is designed to scale across a wide range of compute-intensive markets and applications, such as AI training supercomputing, data centre processing, automotive, metaverse processing, robotics, and life sciences.

• What are some of the new technologies and innovations that you believe will address the shortcomings around energy efficiency, reliability etc?

Ceremorphic will compete against larger semiconductor manufacturers that are creating CPUs and GPUs using the advanced node while being one of the few start-ups with access to 5nm. Individuals who seek to develop electronics with superconductor quantum computers, which utilize particles such as atoms to make machines that can execute jobs that ordinary computers cannot are motivated by the ability to transport electricity in one manner. Because even little amounts of heat might cause quantum computers to collapse, engineers must build them in cryogenic coolers that keep them just above absolute zero. The problem is exacerbated by the fact that standard electronics do not work well at high temperatures. An ultra-cold superconducting diode, on the other hand, may flourish. Conventional computers may also benefit not your desktop or laptop, but rather massive powerhouses like industrial supercomputers. Massive server racks that line the world’s data centers are another potential beneficiary. They utilize 1% of global energy, which is comparable to the consumption of whole mid-sized countries.

We are taking a step in the direction now; Ceremorphic’s Hierarchical Learning Processor, which is billed as an “ultra-low energy supercomputing device,” has a customized machine learning processor, a special floating-point unit, and unique video engines. The company has also created a customized PCIe 6.0 interface. Ceremorphic will profit from this tailored technology in three key areas: reliability, security, and energy efficiency. They will also allow the company to produce chiplet designs for a variety of products ranging from virtual reality glasses to supercomputers.

• Go to market strategy for Ceremorphic in India

Even when Covid-19 was still widespread in 2021, we had the good fortune to successfully carry out our strategy. We produced a ton of IP in the areas of efficient processing microarchitecture, machine learning, and compiler technology. In 2021, we also communicated with possible partners, and this year is no different. Collecting crucial needs and incorporating them into our final product specification was a huge achievement for us. It takes a lot of work to produce something as complicated as QS1. To produce this product quickly and effectively, we are utilizing the skills and abilities of the Indian team. Due to the significance of high-performance computing for the development of AI and machine learning, India would be the primary target market for our go-to-market approach.

We propose to develop end products through our OEM network for target industries including datacenter, AI training, automotive, robotics, metaverse processing, and life sciences. Additionally, using our hardware aftermarket items, workstations, and servers, we will be creating system solutions. The India Development  Centre is already engaged in the development of architecture, algorithms, semiconductor design, and software tools, and it will have a substantial influence on R&D and product development. However, the product will be accessible everywhere. For the businesses in the sector, vertical integration has become essential. Ceremorphic is an ideal partner to several industry titans in their respective fields because of its distinctive technological portfolio and skilled crew.