embedUR-Synaptics: Partnership for the ages!

embedUR systems and Synaptics Inc. have partnered to deliver an affordable, AI-powered edge solutions for India's smart device market.

Michael Hurlston, Synaptics CEO, is a pioneer in Wi-Fi. He was the father of building connectivity in devices. Today, Synaptics is a thought leader in the IoT snd semiconductor spaces. He said that we are lucky to have a good partner in embedUR. We are now seeing the AI-powered edge solution for smart devices. 

Rajesh C. Subramaniam, Founder and CEO, embedUR systems, said that we have seen seismic shift in the amount of compute available at the edge. The hardware has the ability to do intelligent models at the edge. We joined the EdgeAI Foundation a year and a half back. We have ModelNova.AI -- an open platform. It has 40-45 models that can be downloaded anywhere. Custom solutions can be built. By close to end of next year. we will have nearly 150 models. Developers can try them out and build products. Some of these solutions are available today. The chips are small and resource constrained. 

Majority of companies using cloud, their opex is tied to cloud.  AI generally happens in the cloud, driving significant expense.  The innovation that EmbedUR and Synaptics are bringing to the markets is the AI happens on chip at networks edge.  This can be used in homes, factories, enterprises, etc. It is also agnostic. You can also change data depending on the data that you train it with.  Not all apps need to be run on Blackwell, from Nvidia.

This is the first generation of chipsets with intelligence built in. It will create an opportunity for literally everything. Technology is well ahead of use cases. When everything becomes intelligent there is going to be growth of opportunities. Models have pre-trained data. Now, you can optimize devices. Industries can also come in and use the infrastructure. We are bringing AI to life, to any use case, 

We are investing capital, and are hiring actively. We are also upskilling the teams. We want to quadruple investment over the next few years. Products can come out by 2027-28. We have to validate the product. We are also working on the 'Invisible Eye' with Synaptics.

Need to propel concept of edge AI further
Michael Hurlston, CEO, Synaptics, said we have hardware and software for the chip. We also have some models that can compile into the hardware. There are different kinds of models that can be compiled into the hardware. In this way, there is no dependence of the data center. This has lots of significant benefits.  Data centers have extraordinary Energy, power, water, requirements. We can now have a closed system that ultimately reduces the need for data center buildouts, curtailing environmental impact.  We need to propel the concept of edge AI further. That will create ease of use scenario for the semiconductor opportunity in new generation.

Almost every electronic system has an MCU. We are now bringing in AI. Children can get left in cars in the USA, for example. We are enabling low power, low footprint devices. You can detect a child in the car. Another area is industrial, or Industry 4.0, where you can do cycle counts. You can do this based on models from embedUR.

Nvidia sells Blackwell processors for $10,000. We are talking about $5-10! We can do simple use cases. and do tera-ops for use cases. Blackwell capability is 1000x more, and so is the cost. We are trying to bring AI to the masses. It can be game changing for a country like India.

We have a relationship with embedUR. India is emerging as a significant market. The industry has been building their inventories. We are a fabless company. Wafer fabs are being built in Gujarat. There is the packaging piece. India companies are entering packaging,, and we will play a part there. We are also seeing geopolitical concerns, and manufacturing is moving out from China. We have major part of our manufacturing in Taiwan.

If US tariffs are imposed. demand will likely fall. Prices may ultimately go up. We are actively trying to diversify where we supply from. We are manufacturing more in Taiwan, and are now looking at India. We are looking at assembling chips.

The US is currently behind in manufacturing. TSMC is the largest fab player, and is developing another fab in Arizona. India will take some time to catch up. Once the new administration comes in the USA, we will probably have a better picture. 

We are just starting today. This partnership will  be a multi-year journey. The TAM for this market is about $25 billion. All of these will ultimately run on AI. This can lead to further opportunities. We are looking at security and safety systems, industrial, medical, etc. We have opportunities in TV sets, where there could be some objectional content. Factories also have opportunity in predictive maintenance.

Carnegie Mellon has identified different models. You have to train the models to identify any situation. It is a very hard problem to solve. Also, traffic going back to data centers have chances of getting hacked.

Democratizing Edge AI with Astra AI-native platform
Regarding the partnership with Edge AI Foundation, Rajesh Subramaniam, embedUR systems, said that the embedUR systems' partnership with the Edge AI Foundation was still a concept when we came in 1-2 years ago. We realized that we need to take resource constrained environments, and optimize them for intelligence. You need to test a lot. Customers have the capability. They are now looking at the new capability. You can use the intelligence to solve the problems. 

The Synaptics team is already building software with the Astra AI platform. It also has models. We have the width as well. We come in where we can give the breadth of models. These are as good as the data that they are trained on. Every customer also has volumes of data. However, does that data behave and predict the way they should? It needs validation before it becomes a product.

All training has to be done offline. We build the models, and then, combine with optimized software to power the systems.

As for ModelNova for edge AI solutions, Subramaniam added that we have launched ModelNova for efficient models running in various resource constrained environments. Buying GPU, compute, etc., is still expensive. We can download models right away, and check whether it is fit for a product roadmap. The idea is to shorten the lifecycle of product development. We can also learn, as a community. We can help bring a product to the market.

Michael Hurston, Synaptics, said that energy harvesting is necessary. We are trying to reduce power, and run on low-voltage batteries. Ultimately, we would like to do power harvesting. Our goal is to move from corded to battery power, and ultimately, to power harvesting. We work with embedUR systems for this. 

TinyML ideally needs energy harvesting. Subramaniam noted that TinyML brings significant data mining and analytics capabilities, for numerous use cases and verticals. Hurston added that we can generate, analyze, and refine the model each time. We are adapting the model actively, on the fly.