India's semiconductor landscape: a discussion with deep tech startup SandLogic

Prasanth Aby Thomas, DIGITIMES, Bangalore 0

Kamalakar Devaki, Founder, SandLogic. Credit: SandLogic.

The Indian semiconductor industry is developing well, fueled by government initiatives, current geopolitical concerns, and rising private sector interests. These developments have helped retain talent within the country, allowing professionals to work on significant projects without having to go abroad. The central and state governments have launched numerous initiatives to bring talent into the fabless semiconductor space.

However, challenges remain, as the semiconductor industry is capital-intensive and relies heavily on government funding, which can be slow, according to Kamalakar Devaki, Founder of SandLogic, a full-stack, deep tech company that develops solutions and products for both the enterprise AI and industrial AI segments.

"The investment amounts needed are much higher than what is currently available," Devaki suggested. "There is a need for increased private participation, as many Indian investment firms have yet to add the semiconductor industry to their portfolios. Only a handful of firms have a good understanding of the industry."

But despite these challenges, the government's initiatives are bridging gaps in the industry, providing access to tools, fabs, and exposure, Devaki added. SandLogic is part of these efforts, taking advantage of the opportunities offered by the government.

Catering to different segments

SandLogic has been working with customers around the globe in building commercial grade solutions, using machine and computer vision, NLP, and edge AI, all of which fall under deep learning space. Over the last five years, SandLogic has created three different products that cater to enterprise and industrial AI spaces.

"In the enterprise AI segment, SandLogic has two main products: TXTR and Lingo," Devaki explained. "TXTR is an OCR/ICR framework that helps businesses automate use cases involving document digitization, accounts receivable and payable automation, and more. The product goes beyond basic OCR capabilities, enabling complex automation use cases involving engineering documents and even handwritten text."

Lingo is a comprehensive framework for handling speech and audio use cases. This solution starts with voice fingerprinting to identify speakers and extends to automatic summarization of meetings, transcription of calls, and analysis of emotions and sentiments. Lingo has been specifically designed for call centers and customer care centers to perform quality analysis, providing detailed insights on various aspects of inbound and outbound calls.

"On the industrial side, SandLogic has developed a product called Edge Matrix." Devaki added. "This product is a sophisticated framework for running AI workloads on small edge devices with limited RAM and computing capabilities. For example, we recently ported a 21-layer neural network onto a device with just 258 KB of memory. And then we came up with a very detailed framework, which today we call ExSLerate. It's a deep learning accelerator that helps you to run any AI related neural networks on small devices."

ExSLerate has been tested on both RISC-V and ARM architectures and is ready for implementation in ASIC formats. With the ExSLerate DLA IP, SandLogic can help create low-cost AI SoCs that could be based on RISC-V or ARM architectures. As the adoption of RISC-V is increasing worldwide and with the Indian government's association with RISC-V, Sand Logic's DLA is poised for growth in the market.

Demand on the rise

The demand for conversational AI and generative AI is driving significant interest in the market, as seen with the emergence of chatbots and new AI possibilities. SandLogic's products, TXTR and Lingo, align with this trend and have wide applicability across various domains due to their domain-agnostic nature.

"In addition to conversational AI, there is a growing need for edge computing solutions," Devaki said. "This need arises from the desire to perform AI inference at the edge, save on data bandwidth and cloud costs, and protect proprietary data. Major companies like Microsoft, Google, Intel, AMD, and Xilinx have recognized this need and started creating inference hardware to cater to the market."

Specialization in AI has led to the development of customized algorithms and an increasing number of customers choosing edge computing over cloud-based solutions. Market research indicates that by 2025-2030, 75 percent of AI deployments will prefer an ASIC format. Many enterprise customers are already offloading their AI workloads to edge devices, with some planning hybrid scenarios where parts of the neural networks are in the cloud and decision-making or other specific use cases are deployed at the edge.

"These driving factors emphasize the importance of making timely decisions and positioning oneself at the center of this growing trend towards edge computing," Devaki added.

Growing with India's indigenization boost

SandLogic has been actively following and participating in the India governments initiatives since 2019, particularly those led by the Ministry of Electronics and Information Technology (MEITy). The company has taken part in initiatives such as the Microprocessor Challenge, which aimed to promote the usage of RISC-V processors. This challenge invited startups to develop various solutions and was part of a larger startup scheme.

"In the competition, which saw the participation of over 6,300 companies, SandLogic emerged as the top-ranked finalist among the 30 finalists," Devaki said. "So we stand to benefit from the Government of India's schemes in the semiconductor space, especially for the ExSLerate DLA IP and its future developments."