New processor architectures require access to development environments to create and optimize software. This presentation describes a next-generation SoC that enables professional developers to create RISC-V applications from bare-metal to Linux-based, including porting of existing applications. The Freedom U740 next-generation SoC combines a heterogeneous mix-and-match core complex with modern PC expansion capabilities and form factor with a suite of included tools to facilitate broad professional software development.

Mobile computing continues to evolve rapidly. New use cases and form factors are driving aggressive and seemingly contradictory design points. This poses a challenge for companies as they build innovative silicon for tomorrow’s devices. This presentation will provide an overview of how the latest Arm CPUs are unlocking novel, flexible, and holistic solutions. These CPUs help deliver high-performance, efficient, and cross-platform heterogeneous compute without compromising on security or time-to-market.

As AI and ML drive chip complexity, heterogeneous architectures using multiple types of processing elements are becoming a practical solution to meet processing and power requirements in systems-on-chip (SoC). This presentation describes new technology that allows processors developed with AMBA CHI, ACE and AXI interfaces to be integrated together in a single cache-coherent system. The talk includes graphical examples that highlight topology and network node configuration flexibility and the use of integrated system-level simulation to determine optimal architectures.

Standard CPUs, with tens to hundreds of cores and ALUs, can perform complex calculations on small datasets very quickly, but they are less efficient when handling large datasets due to the memory bottleneck imposed by traditional von Neumann architecture.  When a task requires performing parallel calculations across a large dataset, a higher-performance and more energy-efficient “Associative Processing Unit” can be designed as an inter-connected array of millions of simple “boolean bit processors” utilizing in- and near- memory processing techniques.

Traditional processor architectures are failing to keep up with the exploding compute demands of AI workloads. They are limited by the power-hungry weight-fetch of von Neumann architectures and limitations of transistor and frequency scaling. At-memory computation places compute elements directly in the memory array, providing reduced power consumption and increased throughput due to the massive parallelism and bandwidth provided by the architecture. This presentation introduces a new class of non-von Neumann compute designed to meet these AI demands.

Micro edge applications present severe constraints in power consumption and cost. Architectures based solely on off-the-shelf components cannot meet these requirements.  New innovations are required, all the way down to circuits and devices. This presentation discusses the implementation of the GPX-10, a general-purpose AI processor, and its DigAn technology platform, which uses custom circuit components to deliver high-performance AI inference and retraining within the power and cost requirements for micro edge devices.