Joint Effort Focuses on Accelerating AI in Automotive Apps

Artificial intelligence (AI) and neural networks are becoming a key factor in developing safe, smart, and eco-friendly cars. To support AI-driven solutions using its future automotive microcontrollers, Infineon Technologies has initiated a collaboration with Synopsys. AURIX microcontrollers from Infineon will integrate a new AI accelerator called a parallel processing unit (PPU) that employs Synopsys’ DesignWare ARC EV Processor IP.

AI and neural networks are fundamental building blocks for future automated driving applications, such as object classification, target tracking, or path planning. They also play an important role in optimizing many other automotive applications, helping to reduce the cost of electronic-control-unit (ECU) systems, improving their performance, and accelerating time-to-market.

For example, they enable optimized engine auto-calibration and reduce the number of sensors by producing accurate mathematical models of the physical reactions occurring in a system. At the same time, however, AI applications require much higher computing power than standard algorithms.

“By developing the PPU together with Synopsys, we make sure that our future microcontrollers will provide the safety features, throughput, and power-efficient performance necessary to meet increasing AI computational requirements,” said Peter Schäfer, head of the microcontroller business line at Infineon. “This will prepare the AURIX for data-hungry automotive applications such as future gateways, domain and zone controllers, engine management, electro-mobility, and advanced driver-assistance systems.”

AURIX MCUs currently support certain types of neural networks. However, according to Infineon, the PPU will take its real-time and AI capabilities to a new level with significantly higher performance than today’s accelerators, enabling AURIX to process data from advanced sensors where it is currently bounded by factors such as real-time constraints. The PPU will accelerate AI algorithms such as:

• Recurrent neural networks (RNN), which can use internal memory to process sequences of inputs. This makes them applicable to tasks such as handwriting recognition or speech recognition.
• Multi-layer perceptron (MLP), a neural network connecting multiple layers in which the signal path through the nodes only goes one way.  An MLP is a class of feedforward artificial neural network. It consists of at least three layers of nodes: an input layer, a hidden layer, and an output layer. An MLP utilizes a supervise learning technique called backpropagation for training.
• In deep learning, a convolutional neural network (CNN) is a class of deep neural networks most commonly applied to analyzing visual images. CNNs take advantage of the hierarchical pattern in data and assemble more complex patterns using smaller and simpler patterns.
• The value of a radial basis function (RBF) depends only on the distance between the input and some fixed point, either the origin or some other fixed point called a center. Any function that satisfies the property is a radial function. RBF networks have many uses, including function approximation, times series prediction, classification, and system control.

“In many AI-driven applications, safety is paramount,” said Joachim Kunkel, general manager of the Solutions Group at Synopsys. “Combining the processing power and safety features of our ARC EV Processor with the proven architecture of the AURIX will enable the development of automotive systems at the highest levels of functional safety.”

The EV Processor is supported by Synopsys’ MetaWare EV Development Toolkit for Safety, which speeds safety-compliant application software development for automotive designs. The resulting AURIX toolchain will support model-based designs, enabling software design strategies and reducing the increasingly demanding automotive time-to-market.

Furthermore, with its CNN support, the PPU will enable layered security concepts supporting techniques for intrusion detection and prevention systems like deep packet inspections or system entropy monitoring.