Semiconductor veterans back AI targeting physical design, a key constraint for fabless ecosystems

Founded by IIT Madras alumnus Bragadeesh, Tattvam AI is building an AI abstraction layer over EDA tools from Cadence and Synopsys to shorten multi-year chip design cycles.

Semiconductor investors are backing Tattvam AI as it develops what it describes as an AI reasoning layer for chip design.

The company aims to reduce the manual effort required to build custom silicon as global demand for specialised AI processors accelerates.

The deeptech startup has raised $1.7 million in pre-seed funding led by Seedcamp, with participation from EWOR, Entropy Industrial Ventures, Concept Ventures, and semiconductor angel Stan Boland, former founder and CEO of Icera and Element 14.

Boland previously led the sale of Element 14 to Broadcom and sold Icera to Nvidia in 2011.

Speaking to CRN India, Tattvam AI co-founder and CEO Bragadeesh Suresh Babu said the company is targeting the physical design stage of chip development, now emerging as the primary bottleneck in custom silicon creation.

“Physical design converts chip architecture into manufacturable layouts. It is a highly iterative and complex process, demanding thousands of engineers and months of painstaking work. While verification and logic synthesis have seen efficiency gains through AI, physical design remains largely manual, slowing the overall chip cycle,” said Babu.

Bragadeesh founded Tattvam AI with Lannan Jiang, who has been developing chips at a research lab at ETH Zurich.

According to the company, it plans to make custom silicon accessible, reduce development costs, and enable rapid iteration on chip designs.

It aims to add an AI-driven layer on top of existing EDA tools from Cadence and Synopsys, and claims that this will compress chip design cycles from two to three years to weeks.

Faster chip iteration allows startups and service firms to align designs with evolving AI workloads, reducing capital risk and enabling quicker pivots in product development.

Industries like robotics, biotech, and medical devices will benefit from shorter design cycles that previously took years to execute.

Babu said, “Physical design is like a compiler in software. Our AI abstraction layer understands design intent and translates it into manufacturable layouts, enabling rapid iteration without replacing the existing EDA stack. This allows hardware to move at software-like speed.”

Custom silicon becoming central to AI infrastructure

Unlike general-purpose processors, specialised chips are designed for specific workloads such as AI training and inference.

These processors can deliver higher performance and improved power efficiency for targeted applications.

Technology companies are building in-house silicon to optimise performance and control costs. Google develops Tensor Processing Units (TPUs) for AI workloads, while Nvidia partners on inference-focused architectures.

Despite this demand, semiconductor design remains a slow and complex process. Chip development typically takes years and depends on a limited pool of experienced engineers.

While AI systems are already generating software code at scale, their application in chip design remains limited.

According to Tattvam AI, it is building an AI system to understand circuit structures and autonomously solve complex design tasks.

The company says its approach focuses on reasoning over circuit constraints, trade-offs, and interdependencies rather than relying solely on large language models.

Talking about the Indian ecosystem, Babu said it will reduce capital and time risks for companies and startups, improving their ability to compete in fabless semiconductor markets.

By shortening design cycles, startups can better align chip capabilities with fast-evolving AI model requirements. This approach counters current challenges where companies build chips based on outdated market assumptions due to long lead times.

Babu previously worked at UK-based brain-monitoring startup CoMind and was among the early engineers at chip startup Fractile. He chose to launch Tattvam AI instead of joining Google’s TPU team.

Ecosystem-led go-to-market strategy

Babu mentioned that Tattvam AI is building its go-to-market strategy around ecosystem alignment.

The startup is holding exploratory discussions with fabs, EDA vendors, chip design firms and semiconductor service companies to ensure smooth integration into existing workflows, he said.

No formal partnerships have been announced yet.

He told CRN India that integration is essential given the tightly coupled nature of semiconductor development. “Design, tools and manufacturing are interdependent. We are building on top of the existing stack, not replacing it,” he said.

The investor base also provides strategic access and contributes industry feedback and introductions across the value chain.

Babu clarified that while Nvidia is viewed as a potential customer, it is not an investor in Tattvam AI.

It plans to work as an independent innovator building horizontal infrastructure for the semiconductor ecosystem rather than aligning exclusively with any single chipmaker.