The growth of autonomous systems, such as self-driving cars, drones and robots, is driving the development of embedded systems that can process complex sensor data and make decisions in real-time—usually with a tight power budget. As a result, software must be cutting-edge, safety-certified and efficient. For example, today’s embedded developers need SDK support for robust machine learning (ML) and deep learning (DL) frameworks, such as TensorFlow, ONNX Runtime and Caffe. Libraries like these and support for optimizing these models for embedded hardware, are critical to drive product innovation.

Developers working in the automotive and industrial markets often develop in environments where safety is paramount. To maintain a high standard, embedded software engineers implement operating systems across multiple hardware cores to manage tasks and verify predictable, low-latency responses to sensor data. Their entire software stack must be safety-rated and tested against international standards such as ISO 26262 and IEC 61508 to reach ASIL B/ASIL D (automotive) or SIL 2/SIL 3 (industrial) ratings. Embedded software engineers also need a strengthened platform to build safety-critical software, such as certified libraries, development processes and tooling.

Power constraints also present a constant challenge for embedded software engineers developing autonomous systems. For example, in automotive front camera applications, thermal constraints limit a processor's power budget to five or six watts because of the typical placement under windshields, which are subjected to sustained high temperatures. Running inefficient software results in excessive power draw, adding unnecessary heat to the system. Too much heat results in thermal noise and poor image quality. Dynamic software design, low-power modes and power optimization tools are essential for developers looking to keep their embedded designs running efficiently.

Figure 2-1 ADAS embedded software engineers require access to safety-critical software