Unleashing always-on edge intelligence with low power analog AI chip 

We are dedicated to building energy efficient analog AI chips that deliver high performance intelligence at the edge, ensuring real-time decision making without reliance on the cloud. Our breakthrough analog computing architecture enables efficient, always-on AI for devices that demand speed, privacy and unmatched energy efficiency.

 

 

                                               Target Products

  Always-On Sensor AI Chip

  • Micro-watt power consumption

  • Supports 4–8 bit in-memory inference

  • Non-volatile weights (no standby power)

  • Minimal digital logic

  • Sub-1V operation

  AI Sensor Fusion Chip

  • Handles multi-modal signal preprocessing

  • In-memory compute for fusion networks

  • Low-latency analog accumulation

  • Ultra-low-power federated fusion engine

   Analog Neural Co-Processor

  • Ultra-low-power inference engine

  • Integrates as side-core over MIPI/SPI

  • Executes small neural networks for sensing tasks

  • Releases digital NPU from continuous load

  Edge Learning AI Chip (Offline Adaptive Learning)

  • Electrochemical synaptic updates

  • True analog incremental weight tuning

  • Non-volatile memory

  • Local training for personalization

  Vision Edge AI Chip for Drones & Robotics

  • Electrochemical analog MAC arrays

  • Handles small CNNs

  • On-chip feature extraction

  • mW-level consumption vs 1–5W digital NPUs

Our Business

 

  • At Utkalika, our work in Analog Edge AI focuses on developing ultra low power AI chips that compute directly in memory using electrochemical analog processing. This approach eliminates the power-hungry data movement found in digital processors, enabling real time intelligence on devices that operate under tight energy budgets.
  • Our technology supports always-on sensing, in-memory inference, offline learning and non-volatile weight storage, making it ideal for drones, defense systems,smartphones, laptops, industrial IoT, healthcare devices, smartphones and wearables.

Analog Edge AI vs Datacenter AI Chips (GPU/TPU)

 

  • Datacenter GPUs/TPUs deliver high performance but consume 100W–300W+, require cooling and rely on constant internet connectivity.

  • Analog Edge AI operates at micro-watts to milli watts, enabling real-time intelligence without the cloud.

  • GPUs/TPUs are ideal for large scale training; Analog Edge AI is ideal for small, autonomous devices.

 

Analog Edge AI vs Digital Edge AI Chips (NPU/DSP/MCU)

 

  • Digital edge chips still rely heavily on data movement, which accounts for up to 90% of their power consumption.

  • Analog Edge AI performs computation inside the memory cell, virtually removing this energy cost.

  • Digital chips require 8–16-bit precision; analog systems achieve efficient 4–8-bit operations sufficient for edge inference.

  • Analog AI supports offline learning and non-volatile memory, unlike digital NPUs.

 

Contact us

Get in touch with us for your queries 

Location

Utkalika Research and Development Private Limited
Bellandur, Banglore, India-560103

Kendrapada, Odisha, India-754218

About us

Utkalika Research and Development, established in 2024, is a deep tech startup founded by experienced former researchers from DRDO.  Our core team has worked with leading semiconductor companies such as Infineon, Intel, and AMD, bringing together more than a decade of chip design expertise. We are focused on developing next-generation analog edge AI chips that deliver ultra-low-power, real-time intelligence for defense, space, robotics, telecom, healthcare, and a broad range of commercial edge devices. Our team’s strong technical foundation, industry experience and innovation-driven approach enable us to address the most demanding challenges across these sectors with cutting edge, reliable solutions.