Silicon & Systems Research

Indigenous edge-AI silicon, designed with agents

Drutam's research arm builds edge-AI accelerator IP and the system software around it — and pioneers an agentic methodology for silicon development itself.

The IP

The accelerator

Drutam's core IP is an edge-AI accelerator designed around systolic-array compute: INT8/INT16 matrix engines paired with FP32 vector processing. The architecture is prototyped and iterated on datacenter-class FPGAs today, and is engineered for a clean path to ASIC.

Systolic-array compute

INT8/INT16 matrix cores with FP32 vector processing.

FPGA-prototyped

Running on datacenter-class FPGAs; engineered for tape-out.

General tensor substrate

A programmable deep-learning core, not a fixed-function block.

Indigenous IP

Architected, verified, and iterated end-to-end in India.

The methodology

The agentic hardware flow

What distinguishes our silicon R&D is how the silicon is built. AI agents are embedded across the hardware development flow — drafting RTL from specification, generating and triaging verification, exploring synthesis and timing-closure strategies, iterating physical design.

The research output is therefore twofold: the accelerator IP itself, and a reusable agentic methodology for building silicon. We believe this methodology is itself a strategic capability for any nation building a semiconductor design workforce.

The platform

System software research

An accelerator is only useful when models actually run on it. The vertical spans compilers, runtimes, and instruction-set research for resource-constrained edge controllers — so our hardware ships as a usable platform, not a bare chip. The long-term arc is a complete edge-AI deployment platform: silicon, system software, and deployment tooling, designed together.

Continuity

Years in the making

  1. 01

    Origins

    Deep-learning accelerator research begins at IIT Madras with the ShaktiMAAN project.

  2. 02

    Prototyping

    Multi-year architecture iteration; the accelerator runs and is validated on datacenter-class FPGAs.

  3. 03

    Agentic flow

    AI agents embedded across RTL, verification, and physical-design exploration — in production use on our own silicon.

  4. 04

    Ahead

    ASIC realisation and a complete edge-AI deployment platform: silicon, system software, and tooling.

Origin
IIT Madras · ShaktiMAAN lineage
Incubation
IITM Pravartak Technologies Foundation
Honorary Advisor
Prof. V. Kamakoti
Design
Indigenous, end-to-end in India

Made in India 🇮🇳

Indigenous by design

Drutam's accelerator IP, agentic methodology, and system software are architected end-to-end in India — aligned with the national priority of building indigenous semiconductor design capability, edge-AI silicon, and a trained silicon-design talent pipeline.

Institutional & research partnerships

For government programmes, academic collaboration, and strategic partners in indigenous silicon — we welcome the conversation.