🧠 Mythic AI Expands in India

US chipmaker Mythic AI opens a Centre of Excellence in Bengaluru, deploying analog compute-in-memory chips for energy-efficient data centres, robotics, and edge systems.

US-headquartered semiconductor pioneer Mythic AI has inaugurated a premier Centre of Excellence (CoE) in Bengaluru, cementing a major strategic expansion into India's rapidly growing deeptech hardware corridor. The new facility is tasked with end-to-end architecture, physical design, and applications engineering for Mythic's next-generation Vanguard AI inference platform, which utilizes breakthrough analog compute-in-memory (CiM) technology to deliver up to 100 times greater energy efficiency than conventional GPU-based systems.

Under the leadership of Manoj Kumar, Vice President of Mythic India, the Bengaluru center has already assembled a founding team of 20 senior microelectronics engineers, with concrete plans to scale the engineering workforce to over 60 specialists across physical design, design-for-test (DFT), systems firmware, and AI model optimization by the end of Q1 2027.

Breaking the Von Neumann Bottleneck: The Analog Revolution

For over half a century, computing architectures have followed the classic von Neumann paradigm: separate processing units (CPUs and GPUs) connected via high-speed buses to separate memory banks (DRAM and HBM). In traditional artificial intelligence inference workloads, calculating billions of matrix multiply-accumulate (MAC) operations requires continuous, high-frequency data shuttling between memory and logic gates.

This continuous data transfer is disastrous for energy efficiency. In modern digital GPUs, over 80% to 90% of total electrical energy is consumed simply moving numbers across microscopic silicon wires, rather than performing actual mathematical computation—a fundamental physical limitation known across the industry as the "Memory Wall."

Mythic AI circumvents this barrier entirely through analog compute-in-memory:

- Computing with Physics: Rather than converting data into digital binary bits (1s and 0s) and routing them to arithmetic logic units, Mythic stores neural network weights as precise analog electrical conductances directly inside embedded flash memory cells.
- Instantaneous Matrix Math: By applying input voltages along memory rows and measuring resulting output currents along columns, the chip performs massive matrix multiplication instantaneously in the analog domain, leveraging fundamental physical principles (Ohm’s Law and Kirchhoff’s Current Law).
- 100x Energy Efficiency: By eliminating digital data movement between memory arrays and processing cores, Mythic's architecture delivers up to 100 TOPS (Tera Operations Per Second) at a fraction of the thermal footprint and wattage required by standard digital accelerators.

"Data centers and edge systems worldwide are running straight into a hard electrical power ceiling,"

noted Manoj Kumar, VP of Mythic India. "You cannot deploy standard 400-watt server GPUs into an autonomous agricultural drone, a robotic surgical arm, or a congested edge telecom cabinet. Our Bengaluru Centre of Excellence is not a secondary support outpost; it is a primary design engine creating the analog silicon that will power ultra-low-power edge intelligence worldwide."

This hardware breakthrough aligns with sovereign microelectronics momentum across India, connecting with indigenous chip design tools developed by Swadeza's on-premises semiconductor suite and industrial hardware validation scaled by Emerson's 75,000 sq ft testing R&D center in Bengaluru.

Silicon Architecture Comparison: Analog CiM vs Digital GPUs

The structured benchmark matrix below evaluates Mythic's analog compute-in-memory architecture against conventional digital computing paradigms across key performance metrics:

Four Strategic Market Pillars for Mythic India

The Bengaluru Centre of Excellence has targeted four critical industry verticals where electrical power and thermal dissipation represent the primary bottleneck:

1. Power-Constrained AI Data Centers: Hyperscalers and sovereign compute centers—such as those expanding under India's proposed ₹20,000 crore frontier AI initiative—face severe regional electrical grid constraints. Offloading high-frequency inference workloads to Mythic's analog coprocessors enables operators to 10x their inference throughput without upgrading utility substation transformers.
2. Autonomous Industrial Robotics: Enabling collaborative warehouse robots (AMRs) and automated manufacturing arms to execute sub-millisecond 4K computer vision and obstacle avoidance without requiring bulky heat sinks or depleting onboard lithium-ion battery reserves.
3. Automotive ADAS & Driver Monitoring: Delivering real-time multi-camera sensor fusion, cabin gaze tracking, and radar processing within the strict thermal and functional safety envelopes demanded by electric vehicle manufacturers.
4. Defense, Aerospace & Tactical Edge: Deploying ruggedized, low-power neural processors into autonomous search-and-rescue UAVs, persistent surveillance arrays, and border reconnaissance equipment operating under harsh environmental conditions.

Leveraging India's World-Class Analog Microelectronics Talent

While India has long been celebrated for digital RTL verification and software development, the country also possesses a concentrated, highly elite fraternity of analog and mixed-signal integrated circuit designers. Designing analog circuits requires deep mastery of physical silicon behavior, parasitic capacitance modeling, and process variation tolerance—skills honed by Indian engineers across decades of RF and power management design.

By establishing an end-to-end product mandate in Bengaluru, Mythic AI is capitalizing on this unique engineering depth. The Bengaluru CoE is positioned to play an instrumental role in proving that analog computing is not an experimental curiosity, but the defining architecture for the next era of sustainable, energy-efficient artificial intelligence.

Frequently Asked Questions

What is Mythic AI and what technology does it develop?

Mythic AI is a US-headquartered semiconductor pioneer that specializes in analog compute-in-memory (CiM) microprocessors. Rather than using conventional digital logic, Mythic's chips perform artificial intelligence matrix calculations directly inside flash memory cells, dramatically reducing power consumption and eliminating the traditional memory wall bottleneck.

What is the mandate of Mythic AI's new Centre of Excellence in Bengaluru?

The Bengaluru Centre of Excellence is an end-to-end product development hub that holds core ownership for Mythic's next-generation Vanguard AI inference architecture. The facility's mandate spans analog circuit design, physical layout, design-for-test (DFT), firmware compilation, and edge applications engineering.

How does analog compute-in-memory deliver 100x energy efficiency?

In conventional digital GPU architectures, up to 90% of total electrical power is wasted shuttling weights and activations between separate memory chips and processing cores (the von Neumann bottleneck). Mythic stores neural network weights as variable electrical conductances within analog flash memory, executing multiply-accumulate operations in parallel via Ohm's and Kirchhoff's laws at near-zero data movement cost.

Which industries in India and globally are targeted by Mythic's hardware?

Mythic is targeting four high-growth markets: AI data centers facing severe power grid constraints, industrial robotics requiring real-time sub-millisecond edge vision, autonomous vehicles (ADAS), and defense security platforms such as autonomous drones.

Primary Sources & Official References

- Mythic AI Inc. Corporate Technical Whitepaper: Analog Compute-in-Memory Architecture for Next-Generation Inference: Technical documentation detailing flash memory conductance programming and analog MAC pipelines.
- India Electronics and Semiconductor Association (IESA): Global Semiconductor Centres of Excellence in Bengaluru: Industry census on multinational semiconductor design centers and mixed-signal engineering talent.
- IEEE Journal of Solid-State Circuits: Survey of Analog Flash-Based Matrix Multiplication Accelerators: Peer-reviewed academic benchmarks evaluating analog compute-in-memory efficiency and signal-to-noise ratios.
- Ministry of Electronics and Information Technology (MeitY): Semiconductor Design Infrastructure & Bengaluru Engineering Ecosystem: Government policy briefs covering international deeptech design center investments.