TLDR: EdgeQ, a Silicon Valley-headquartered startup with a significant engineering presence in India, is developing the country’s first unified 5G and AI System-on-Chip (SoC). This innovative chip integrates the entire 5G stack with AI acceleration, aiming to disrupt the wireless infrastructure market by offering significant cost and power advantages over traditional multi-chip solutions.
EdgeQ, founded in 2018 by Qualcomm veteran Vinay Ravuri, is challenging the established wireless infrastructure market by developing India’s first unified 5G and AI System-on-Chip (SoC). Ravuri observed a lack of innovation in the wireless chip market, dominated by a few players, and identified RISC-V, an open-standard instruction set architecture, as a catalyst for change, akin to Linux for software. He stated, “I looked around and saw that when it came to wireless, there was one dominant company and then a few that I could count on my fingers.” He added, “RISC-V opened up hardware development. We were among the first to recognize that this could reduce cost and time to market. Today, almost all machine learning companies use RISC-V.”
The company’s global talent acquisition strategy led to a strong engineering presence in India, with teams in Bengaluru, Pune, Noida, and Hyderabad, complementing its US and European operations. Hariprasad Gangadharan, VP and Head of Silicon Engineering, and Country Head India, highlighted the deliberate choice of many engineers to drive advanced technology development from India.
EdgeQ’s core innovation is an SoC that integrates the complete 5G stack (Layer 1, 2, and 3 software, plus digital signal processing) with AI acceleration. Gangadharan proudly noted, “We’re one of the very few startups globally, and the only one from India, building a full 5G SoC from scratch.” This architecture unifies 5G baseband and AI matrix computations using configurable multipliers. Gangadharan explained, “The same multipliers that perform channel estimation and equalization in 5G can also perform neural-network computations.” Ravuri further clarified the technical flexibility: “Our chip architecture allows it to handle both 5G and AI computations. The same matrix mathematics used in 5G, based on complex numbers, can also be used for AI by modifying the elements of the matrix. By zeroing out the imaginary part, we turn it into a dot-product operation, which is what most AI computations rely on.” This software-configurable design enables adaptability across various deployments, including indoor/outdoor small cells, distributed units, and satellites.
The SoC incorporates Tensor Execution Units (TXU) for mathematical operations, Forward Error Correction blocks compliant with 3GPP standards, security and crypto accelerators, custom instruction sets for packet processing, real-time scheduling, ARM CPUs, and RISC-V cores for software programmability. This integration results in substantial cost and power efficiencies. Gangadharan emphasized, “Traditional solutions use at least three chips… Our single-chip approach eliminates this redundancy, drastically reducing both power consumption and the bill of materials.” EdgeQ has achieved “roughly one-third the power of comparable solutions” by applying mobile SoC design techniques to base-station chips.
EdgeQ prioritized manufacturability and scalability. The company achieved “zero metal-mask fixes post-tape-out” through rigorous verification processes, including “extensive modeling, emulation, and software-hardware co-verification,” utilizing SystemC models, real traffic pattern stress-testing, and FPGA-based emulation. The customer journey involves 3-6 months of evaluation using kits, followed by approximately 12 months of system development, with EdgeQ providing ongoing support and firmware updates for the chip’s 5-7 year lifecycle. Customers can also integrate their own Layer 2/3 software via standard APIs and Small Cell Forum FAPI specifications.
The business model focuses on high-volume wireless infrastructure segments like indoor and outdoor small cells, open RAN deployments, and satellites. Revenue is generated through chip sales bundled with base software, with additional licenses for features such as AI functionality or simultaneous 4G+5G operation. Ravuri confirmed, “AI is an add-on feature. Customers who want AI capability pay extra for that functionality.” EdgeQ serves high-end global OEMs and device makers, partnering with cloud and software firms like Radisys (part of Jio) and supporting AWS and Google Cloud platforms. The company has raised $126 million across three funding rounds, including a $75 million Series B in April 2023, with investors like Threshold Ventures and an advisory board featuring former Qualcomm executives Paul Jacobs and Matt Grob. EdgeQ operates on a full Customer-Owned Tooling (COT) model, collaborating with Synopsys, Cadence, and TSMC.
Also Read:
- Indian Deeptech Startup Leverages Generative AI to Revolutionize Analog Chip Design
- On-Device AI Reshapes Consumer Electronics Landscape, MediaTek at the Forefront
Despite facing challenges typical for a disruptor in a market dominated by incumbents, such as slower-than-expected 5G market growth, EdgeQ sees opportunities in private 5G networks, industrial automation, and geopolitical drivers for local manufacturing, particularly in India. The company plans to evolve its unified 5G + AI platform towards 6G and advanced AI integration. Ravuri concluded, “Every disruptive technology takes longer than expected, about 10 years, but once it takes off, it grows quickly. We are at that turning point now.” India is set to become EdgeQ’s primary hub for design, verification, and software, with plans to expand production and AI capabilities on the same chip.
