HYDERABAD: Computers get smarter, faster and more energy efficient with each improvement made to processors and the technology that powers them. But, to meet the computing demands of Artificial Intelligence, computers need more processing power. As a result, traditional processors do not have the ability to process big data, which requires neural network processing. Only a few technology majors across the world, like Intel for instance, have come up with their own family of neural network processors.
However, for the first time, a Hyderabad-based start-up is trying to become the only Indian company to break into the big league of blue chip processor manufacturers by providing a similar solution. Manjeera Digital Systems, incubated at the Centre for Innovation and Entrepreneurship (CIE) in IIIT-Hyderabad, has designed a processor by changing their approach to computing. The five-year-old start-up has designed a processor called the Universal Multifunction Accelerator or UMA that focusses on Research and Development related to Digital signal Processing.
“Unlike other players in the field who have been tweaking and making incremental changes or modifications to the processor while keeping the fundamental architecture at the core the same, we at Manjeera took a step back and created a fundamentally new approach to computing. We call this Universal Multifunction Accelerator or UMA,” said Venu Kandadai, Co-Founder and CEO of Manjeera Digital Systems.
The team tested their high-performance computing engine against existing products in the market for vital parameters such as Performance, Power Consumption, Programmability and Silicon Area. The start-up claims it’s the only one in the market that has achieved desired results of High (Performance), Low (Power Consumption), High (Programmability) and Low (Silicon Area). They also conducted a performance analysis on standard benchmarks for relevant application areas such as deep learning, image processing, computer vision, and DSP to find their processor outperform other leading competitive products.