CEO Interview: GP Singh from Ambient Scientific

Gajendra Prasad Singh, also known as GP Singh, is a seasoned tech professional with over 26 years of experience in advanced semiconductor chips. With a zeal to solve the most complex technical problems, he harped on a difficult journey to create programmable AI Microprocessors, that provide high-performance in a cost-effective manner while still consuming low power. To realize this vision, he co-founded Ambient Scientific along with a team of visionaries from California’s Silicon Valley. GP’s extensive technical experience and successful leadership record within global prestigious companies from building cutting-edge chips contributed to his deep understanding of not only the scientific first principles required for such breakthrough innovations at the grassroots level but also the business acumen to ensure practical feasibility. With an innate passion for everything electronics and computers, GP Singh is a fierce advocate of using semiconductors for the betterment of human lives.

Tell us about your company?

Ambient Scientific is a fabless semiconductor company born in Silicon Valley, pioneering ultra-low power AI processors that are fully programmable to enable endless AI applications.

Our breakthrough Analog In-Memory Compute technology called DigAn® is making AI computing more powerful and efficient than ever before, without compromising on flexibility and programmability. Compared to traditional AI hardware, our processors deliver thousands of time more AI performance at the same power consumption or thousands of time less power consumption for the same AI performance.

Our first product GPX10 leverages the DigAn® architecture to bring battery-powered, cloud-free, on-device AI applications to life, something considered nearly impossible before. From always-on voice detection to FaceID to predictive maintenance, GPX10 is enabling endless applications in various industries, all while running on as little as a coin cell battery with no dependence on the cloud or an internet connection.

With a full stack SDK designed to support industry standard AI frameworks (Tensorflow, Keras, etc.) and an AI compiler to enable custom neural networks, we enable rapid time to market for your AI applications. Order our DVK today and bring the power of AI away from the cloud, right on to your fingertips.

What problems are you solving?

While the AI application and software landscape has exploded in complexity, hardware has failed to keep up. Current chips used for AI processing (GPUs) were designed for graphics processing and not AI computing in mind, making them inefficient and extremely expensive. This is clearly visible with the rising compute costs as well as power consumption for all AI ranging from gigantic LLMs to edge AI for smaller electronic devices. We at Ambient Scientific have solved these problems by inventing not just analog in-memory computing but also new instruction set architecture designed specifically for AI computing. Our analog matrix multiplication engines deliver 40X AI performance at 70X lower power consumption compared to equivalent GPUs. Built with scalability and flexibility in mind, our architecture enables AI processors all the way from cloud and server level to MCU level for a wide variety of applications across several industries. Ambient Scientific’s mission is to make AI computing powerful, energy efficient and affordable for everyone alike.

What application areas are your strongest?

Our first product GPX10 is an AI processor targeted towards on-device AI applications for the tiniest of battery powered devices. It helps move AI processing from the confines of the cloud directly on to the device even if its running on as little as a coin cell battery. This improves application reliability, latency, data security as well as total cost of ownership. Some of our strongest application areas popular with customers are industrial predictive maintenance at the edge, anomaly detection on MedTech devices and cloud-free voice control for consumer products. While commonly these application would struggle with latency or reliability or miniscule battery lives due to AI processing, our processor solves all these problems without forcing any compromises or even affordability.

What keeps your customers up at night?

With the widespread utility of AI, product makers have realized the importance of incorporating AI features into their product roadmap to remain competitive and maintain differentiation. These products makers are now faced with a difficult choice:

1. Run AI processing in the cloud and sacrifice latency, data privacy and reliability due to complete dependence on a network connection.
2. Run AI on device and sacrifice accuracy and power efficiency which translates into significantly compromised battery life.

These limitations which ultimately translate into higher costs or compromised product quality are an absolute function of the current processors available in the market, none of which were designed for AI processing. They force debilitating sacrifices for the product maker that keeps them up at night, stuck between a rock and a hard place.

What does the competitive landscape look like and how do you differentiate?

The AI compute market for small electronic devices includes either MCUs, entry level GPUs or new age NPUs. While MCUs cannot deliver enough performance required for meaningful AI compute, entry level GPUs consume too much power, occupy too much area and are not affordable enough to fit within the boundaries of commercial viability for battery powered on-device AI applications. Several new age NPUs claim to be able to deliver low power AI solutions but with a heavy price to pay in lack of programmability. They tend to be fixed function with pre-defined neural networks and minimal room for customization. Our Ultra-low power AI chips not only deliver the highest performance/unit of power consumption (>7 TOPs/W), they’re smaller than a fingernail, affordable and most importantly completely programmable. Product makers care about programmability so they can differentiate their products from competitors’ by owning the software such as their proprietary AI algorithms. Programmability also makes their products future proof with the ability to push updates over the air as the software and application landscape evolves. Compared to fixed function or application specific NPUs, our processors offer a versatile and flexible platform for product makers to differentiate themselves with ultra-low power AI features well into the future.

What new features/technology are you working on?

Our claim to fame is a breakthrough on analog in-memory computing technology that enables us to leverage a combination of high speed digital and analog circuits designed specifically for AI computing. By leveraging cubic in-memory architecture and the analog matrix multiplication circuit, we’ve solved all the bottlenecks for AI computing while minimizing energy consumption to a fraction of contemporary architectures. Not only this, we’ve also created custom instruction set architecture from the ground up to enable flexibility and scalability in AI computing. This means we can build a wide range of processors from AI MCUs to high speed computer vision processors. Similarly our end to end software stack scales with our processors to adapt to the application needs of software developers for a wide variety of applications in several industries.

Also Read:

CEO Interview: Ollie Jones of Sondrel

CEO Interview: Dr. Yunji Corcoran of SMC Diode Solutions

CEO Interview: Rajesh Vashist of SiTime

CEO Interview: Dr. Greg Newbloom of Membrion

Share this post via:

• 
• 
• 
•