The NPAC executes a run-to-completion vector packet processing (VPP) model that breaks networking tasks into a series of functions or steps called nodes. Each node is assigned to a hardware accelerator or RISC-V core for processing by the scheduler. When a node completes its assigned task, it signals back to the scheduler, which, along with the RISC-V cores, orchestrates the next series of tasks.

Breaking up networking processing into a series of independent nodes has several advantages over processing a single packet from start to finish. Advantages include lower latency, higher throughput, parallel packet processing, separation of control and data planes, and support for modular programming through a library of predefined networking functions.

Typical NPAC applications include IP, layer 4, and VLAN routing, gateway (for example, CAN to Ethernet network bridging), packetization and de-packetization, packet encryption, decryption, authentication, and firewall services. Many of these networking applications can be executed completely independent of the SoC processing cores.

The NPAC connects to the Ethernet switch, PCIe controller, CAN controllers, and SoC memory through a dedicated low-latency packet bus. Dedicated hardware IP blocks handle all traffic flow independent of the SoC cores, eliminating interrupts and minimizing packet latency.

The NPAC uses an intuitive graph-based programming GUI to configure operation. To program the NPAC, the user selects from a list of pre-defined networking functions (nodes), linking them together in a linear fashion. Each node can be easily configured, and new, custom nodes can be added to the library, allowing the NPAC to support custom networking applications. For advanced applications requiring more precise programming, the NPAC can be programmed directly in C.