Article complete

IISc develops design framework for next generation analog chipsets for AI applications

The researchers have outlined their findings in two pre-print studies.

The Indian Institute of Science building surrounded by trees under a cloudy sky.

Key takeaways

  • IISc researchers have created a design framework for advanced analog chipsets aimed at AI applications.
  • The prototype chipset, ARYABHAT-1, can enhance tasks like object and speech recognition with high-speed parallel computing.
  • The findings are documented in pre-print studies and patents have been filed for future commercialisation efforts.
  • Analog chipsets offer advantages in power consumption and size compared to traditional digital chips, despite design challenges.

Takeaways drafted with AI from the article text.

Indian Institute of Science (IISc) researchers have developed a design framework to build next-generation analog computing chipsets. Using this design framework, they have built a prototype of an analog chipset called ARYABHAT-1 (Analog Reconfigurable technologY And Bias-scalable Hardware for AI Tasks). IISc said that this type of chipset could be very helpful for AI-based applications such as object or speech recognition and those that require massive parallel computing operations at high speeds..

The researchers have outlined their findings in two pre-print studies (presently under peer review). They have also filed patents and intend to work with industry partners to commercialise the technology.

Image: IISc

Most electronic devices use digital chips because the design process is simple and scalable. Chetan Singh Thakur, Assistant Professor at the Department of Electronic Systems Engineering (DESE), IISc, said that the advantage of using analog is massive, especially in terms of power usage and size.

  • Testing and co-design analog processors are difficult.
  • Analog chips don’t scale easily. So, they have to be individually customised while transitioning to the next generation technology or to a new application. This makes their design expensive. 
  • The trade-off between precision and speed with power and area is not easy when it comes to analog design.

The researchers said that different machine learning architectures can be programmed on ARYABHAT and can operate robustly across a wide range of temperatures. The architecture is also “bias-scalable.” Its performance remains the same when the conditions like voltage or current are modified. 

Published

Spotted an error? Tell our editors. Read our editorial standards and corrections.

About the author

Contributor

All stories
Belamy · Daily

The day’s AI news, before your first meeting.