Blumind is pitching a farsighted approach of neural network processing using an all-analog architecture for AI applications to be more sustainable in the future. The company has completed the latest phase of R&D and is now looking to commercialize its high-performance, low-cost, and sustainable AI chips.
Instead of building a digital compute architecture like most other industry players today, the company has invented an all-analog neural network signal processor architecture that is totally event-driven, highly efficient and accurate.
The semiconductor industry has started working on reducing the mind-boggling power consumption and improve the sustainability of AI, said Niraj Mathur, co-founder of Blumind. For AI to be more broadly adopted in all aspects of our lives, it has to become more energy efficient, he added.
Having worked in the chip industry for decades and witnessing the surge of AI applications, Mathur and his team at Blumind believe that it is necessary to deliver new innovative semiconductors to enable AI for everyone, everywhere. Markets ranging from smart wearables, health monitoring, to smart mobility and robotics, will demand ultra-low power, low latency and low-cost solutions to become major drivers of global economic growth.
In order to meet the growing demand for AI-enabled applications in people's everyday lives, such as camera-based collision avoidance, microphone-based voice command detection and natural movement for robots, the chip industry has been seeking ways to reduce costs and power consumption of AI chips.
Blumind's all-analog neural network processor can reduce power by 100x-1000x versus current AI chips, said Mathur. The architecture does not rely on absolute voltage or current and does not use ADCs or DACs in the neural network core. This makes Blumind's products impervious to process, voltage and temperature variations without compromising accuracy. Additionally, Blumind's architecture does not require the latest semiconductor fabrication nodes to manufacture their chips, but instead utilize cost-effective mature and readily available fabrication nodes. There is also a trend in the industry of using mature processing nodes due to heightened demand for automotive chips and IoT electronics.
When Mathur met John Gosson, who is now the CTO of Blumind, he knew that AI sustainability will be the next inflection point of the industry. Gosson was already working on the project and the two decided to form Blumind to commercialize Gosson's invention.
Blumind will present its novel processor at TIE Awards in Taipei on Oct 12-14, 2023, aiming to look for lighthouse customers and to build a significant, long-term presence in Taiwan.
The Canadian company closed their latest round of fundraising at the beginning of this year. It is expecting to start another round of fundraising later this year.
Blumind's prototype test board