The powerful general-purpose chips such as CPUs are not in a position to support sophisticated heavy tasks of machine learning and deep learning - a central part of tech giant's operations. As a result, AI chips with parallel computing capabilities are in high demand, and McKinsey predicts that this trend will continue. So now, who will meet this higher 'AI Chip' demand?  

The names of big leading players, including Nvidia, IBM, Intel and AMD, flash across one's mind. Imagine the availability of world-class engineers, sound research background and other resources that rests with these giants.  

Amid this huge but heavily contested market, a startup marked its entry. Founded in 2020, the startup Ceremorphic Technologies Pvt. Ltd., with its office in Hyderabad, Telangana, is designing an ultra-lower power supercomputing chip built-in TSMC 5nm, leveraging its own patented technology and multi-thread processing architecture. We reached out to Dr Venkat Mattela, Founder & CEO of Ceremorphic, to understand their work and their differentiator in the market. 

"Our expertise lies in sustainable differentiation, and this is where we stand out," said Venkat in an interview with INDIAai via Microsoft Teams. An alumnus of IIT Hyderabad and Harvard Bussiness School, he is the author of over 100 patents. 

The Foundation  

The startup is not envisioned over dinner or in a night. It takes a lot of time to build a foundation and then carry the legacy ahead. From 2012 onwards, as per Venkat, there was a transition - it came to the notice that AI is doing things as good as human beings do. Suddenly, there was a huge demand as AI and machine learning require a lot of computing. "So, that was a motivation for me to work on developing supercomputing chips," said Venkat.  

Venkat built his last startup - Redpine Signals, way back in 2006, and it took over 14 years of toilsome work. In March of 2020, the company's low-power Bluetooth and Wi-Fi chip technology (against the offerings by giants Qualcomm and Broadcom) was sold to chipmaker Silicon Labs for a whopping $314 million. The chip is now utilised in the recently released Garmin Fenix 7 smartwatch.  

"The chip I designed consumes 26 per cent lower energy when compared with the leading players in the market," said Venkat. 

After Redpine, Venkat went on to retain 18 employees from Redpine and kept his advance toward chip-making efforts in stealth mode for two years before officially announcing the entry of Ceremorphic. 

Building on the foundation 

For Venkat, the sustainability differentiator is the key; with the same logic, he is building Ceremorphic. The company has already filed over 100 patents. Talking about the system, Venkat says that the team is working on a system with high reliability and low energy demand. As the demand for computing is low, it eventually results in higher processing. 

• When it comes to high reliability, errors cannot be eradicated, but the team assures us that the system must not fail. Imagine a project being run on a supercomputing device, and the system fails on the 29th day; the process has to repeat from the beginning. This needs to go. 
• Secondly, the chip will have a low energy demand. Anyone working on a machine learning programme has to go through the training phase that needs a device with supercomputing performance, which is both time consuming and costly affair. With the chips from Ceremorphic will have a much lower energy demand, the power of supercomputing performance can be brought to the individual computers of developers and engineers. 

The capabilities on which the team is working have the potential to make AI and machine learning pervasive.  

Challenges at hand 

"The talent pool in India is good enough for working in the chip architecture domain, so having a good talent pool is not the challenge but since multinational companies are competing for the same talent pool, retaining this talent pool is a bigger challenge," said Venkat.  

The next big thing comes in the field of research and Development (R&D), where the Indian talent pool is very good. However, the problem starts when it comes to execution or finally releasing the products. For Ceremorphic, both R&D and product development happen at their Hyderabad office in India. 

Have a long-term vision 

With over 30 years of experience in semiconductor products, Venkat rightly pointed out two vital things that need to be addressed. 

• First, it should be very clear to understand that chip designing is not a few months process; that mindset needs to be changed as it's a very long-term process and takes years to design and finally have a product ready.    
• Second, a huge amount of money must be at the disposal of companies working on designing and producing a chip, without any disturbance in between. 

"Depending on the complexity, for a chip, the design time itself is three years, production time is two years - to get the first dollar, one has to wait for at least 5-6 years," said Venkat. This is a different game compared to the software industry, where one can make money from a product in even six months. 

Chip designing and production require a minimum of four years. "That's where most of the companies fail; the team is good, and strategy is there, but lack of long-term funding is one the biggest hurdle," observed Venkat.  

"At Ceremorphic, I have decided to invest the fund myself for four years. So, till 2024 we have our own funding," he concluded.