Specialised chips are essential to deploy AI models on the edge. Yet, most of the chips currently available in the market fall short of these requirements. A few companies across the globe, like Sima.ai and Hailo, are trying to fill that gap.
However, Netrasemi, a Thiruvananthapuram-based Indian startup, is now entering the space with its cost-effective and power-efficient advanced AI chipsets for edge AI use cases.
Backed by the Ministry of Electronics and Information Technology (MeitY), Netrasemi has developed three chips so far to run AI models on the embedded edge. Soon, this startup will tapeout two chips based on the 12 nm nodes.
The first chip – Netra A4000 – represents a line of high-performance chips designed with advanced Chiplet (D2D integration) technology. These chips, offering between 32 and 100 TOPS (trillion operations per second), are targeted at the edge server market, smart NVR systems and robotics industry.
The second and third chips – NetraA2000 & NetraR1000 – are designed for surveillance, video analytics, vision processing, smart cameras, and smart home use cases.
The NetraA2000 is based on an Arm architecture and provides performance of 4 TOPS, whereas the NetraR1000 is based on a RISC-V architecture and offers 2 TOPS performance.
According to Jyothis Indirabhai, CEO and co-founder, Netrasemi is the first startup in India to develop AI chips. Founded in 2020 by Indirabhai along with Sreejith Varma and Deepa Geetha, the startup is a beneficiary of MeitY’s Design Linked Incentive (DLI) scheme.
It was also one of the first startups to be approved for the government’s Chips to Startups (C2S) programme.
( Jyothis Indirabhai receiving an award from Rajeev Chandrasekhar)
Tapeout Soon!
The startup plans to take NetraA2000 & NetraR1000 chips, which will tapeout soon, to customers this year.
“We plan to showcase these products to our customers this year. Our goal is to introduce them to our clients first, allowing them the necessary time to validate the products and ensure they meet production standards,” Indirabhai told AIM.
Over the past several months, the startup has engaged with key OEMs in sectors such as telecom clients, industrial products and surveillance products makers.
“These are just one or two large global players in their sector. Our conversations with them are deep where they have started sharing specifications as per their requirement,” Hariprasad C, the chief strategy officer of Netrasemi, told AIM.
The startup has also formed a partnership with a US-based chip company (name not disclosed) to jointly develop and introduce another version of AI chip to the market.
According to Indirabhai, the startup’s chips are fabricated by Taiwan Semiconductor Manufacturing Company (TSMC), and will hit production by 2025. “The customers will be able to get the product ideally somewhere by mid-2026.”
It’s All About Software
Many chip companies in today’s age project themselves as a software company which builds their own silicon. NVIDIA might be the king of GPUs- but its moat is CUDA- its general purpose parallel computing platform and application programming interface.
Netrasemi also offers a versatile SDK called NetraSDK that supports TensorFlow, PyTorch, and ONNX frameworks.
The SDK enables developers with minimal coding experience to integrate their AI models and applications onto the Netrasemi platforms using an intuitive graph builder tool.
“Rather than simply providing an SDK and expecting users to develop based on that, our framework is far more comprehensive, featuring a wide range of sample applications.
“In many instances, we deliver near-complete products to our customers, allowing them to refine and finalise the solution. Many of our designs are nearly ready for production,” Indirabhai said.
Competition with NVIDIA?
With the Netra A4000, the startup will enter a space where NVIDIA also operates in. NVIDIA’s Jetson Orin series is targeted at the high-end edge AI device market.
However, Indirabhai points out that the A4000 chips do not cover the whole spectrum of applications that Jetson Orin does. But, on the applications the A4000 chips target, Netrasemi manages to offer higher performance.
“For example, while Jetson Orin might advertise 32 TOPS, this figure doesn’t necessarily reflect video analytics performance. Effective video analytics demands specialised IPs tailored for the pipeline and Jetson Orin’s general-purpose architecture isn’t optimised for this.
“Although Jetson might offer 32 TOPS for AI, the actual application performance for video analytics may fall short. Our architecture is superior and offers 10X times more power efficiency,” he pointed out.
The higher-end version of the Jetson Orin series does offer over 200 TOPs, which are meant for high-end compute and edge AI use cases. Here, it is important to note that Netrasemi’s roadmap sees them entering the AI server market sometime in 2027-28.
( Netrasemi lab)
Support from MeitY
As part of the government’s DLI scheme to support semiconductor companies, Netrasemi is bound to receive financial aid of around INR 15 crore. So far, the startup has already received around INR 2.5 crore as refunds.
Moreover, as part of the C2S programme, the startup has received additional support of INR 5 crore.
“One is the DLI, a reimbursement programme spanning three years, and the other is the C2S which is a grant received for the R1000 chip. This grant is part of a joint project submitted in collaboration with the College of Engineering Thiruvananthapuram (CET),” Hariprasad revealed.
Additionally, the government is also supporting the startup with access to electronic design automation (EDA) tools for three years. The startup has also raised $1 million in equity funding from venture capitalists. Indirabhai said they plan to raise funds again in the coming months.
Made-in-India Chips
The emergence of Netrasemi comes at a time when the government of India has significantly pushed for ‘Made-in-India’ chips. Other companies, such as InCore Semiconductor and Mindgrove Technologies, have or are planning to launch their own chipsets in the market.
For instance, Mindgrove has developed its own indigenous microprocessor chip named MG Secure IoT. However, taking these chips to the market remains the biggest challenge.
Undoubtedly, these companies, including Netrasemi will face intense competition from mature players in the market as well as Chinese companies providing cheaper alternatives.
