 Good morning, my name is Marcus Meyer. I am product marketing manager in the microcontroller group with ST micro electronics today I'm going to show you Our new product the SM32 H7 dual core device as well how it works together with our SM32 cube.ai tool which allows you to convert free-drain neural networks into code running on the microcontroller the SM32 H7 SM57 is a new product in our SM32 microcontroller family It has two microcontroller cores on it It has one Cortex M7 core running up to 480 megahertz and it has as well a secondary core and Cortex M4 core running up to 240 megahertz So it comes together with a lot of peripherals. It has USB to High-speed USB on the go several New York's I square C's graphical interface as well as 2 megabyte of flash and up to 1 megabyte of RAM One of the other interesting features of this family is that it works together very well with our new development tool the SM32 cube.ai What is the SM32 cube.ai? So the SM32 cube.ai is a tool which allows you to Take a pre-drained neural network and convert it into code which can run on the microcontroller So What what what does it do? So you have different steps in a process to develop a neural network. So you need to capture the data You need to clean up label the data build your neural network And then when you have done those steps Then you use our tool the SM32 cube.ai Convert it into code then your process analyze it and test the code directly on on hardware So in this specific example demo what we are showing here is we are using an SM32 h7 Discovery board which has a nice display and we also have a VGA camera on back and We are basically taking pictures of this tablet here which displays pictures of food and We're using a neural network which basically can detect what kind of food is displayed So you see here for example in the example a cappuccino and you see how our software using a neural network How it can detect the different kinds of foods so What we've done here we use the neural network based on the FD mobile net technology from a public Domain we use that data set in the public domain then we implemented the cameras So we continuously we can continuously take pictures or do a one-shot mode and Then we use either floating point or mix floating point fixed point algorithms to basically compute the food so Just also to give you an example What the our tool does in terms of footprint so in this specific examples? We're using a 205 kilobyte of RAM and 190 kilobyte of flash and so with our device Which is running at 400 megahertz the results. We're getting Running at 400 megahertz. We're getting 6.2 megahertz so 150 milliseconds per inference Which gives you about six frames per second and an average accuracy of about 80% To summarize we're using an sdm32 h7 discovery board with our dual-core part With a display so then we're using a camera on the back taking the pictures of the food Running the food detection algorithm and presenting the result and the accuracy here on the screen If you want more information about our new sdm32 h7 dual-core Microcontroller as well as our sdm32 cube.ai Tool, please go to our website at sd.com slash sdm32