Why is it difficult?

Moore’s law guided the growth of integrated circuits over the last 50 years with the production of more integrated and low-cost electronic devices. Now it doesn’t work that efficiently, and we no longer get the same power efficiency that we used to get by making smaller transistors. Therefore, new approaches are required to scale computing in the future. In this direction, new technologies such as on-chip parallelism, heterogeneous and specialized parallelism hold a good potential because it is a more power-efficient and scalar way to get performance scaling than general-purpose processors. However, specialized heterogeneous parallelism is hard as we have lost much of the durable, long-term abstractions that we had between hardware and software for 50 years. We are programming in a very different way than we used to, and we have lost the ability to port software as easily as we used to, and that has implications on reliability and security. Therefore, developing a robust hardware-software stack with heterogeneous computing is an open challenge.

What is the impact?

Over the last many years, we have seen that our laptops and cell phones are getting cheaper, smaller in size, and more powerful than before. We expect the same trend to continue in the future. Heterogeneous computing is also energy efficient as it uses different types of processors incorporating specialized processing capabilities to perform different tasks.