Designing a cost-conscious DDR-less EtherCAT slave means fitting communication firmware, protocol functions and application software into a constrained memory footprint without sacrificing predictable operation. The AMIC110 reference design demonstrates one approach: run the EtherCAT slave stack using the SoC’s internal memory, with programmable real-time unit industrial communication subsystem (PRU-ICSS) cores handling time-critical data-link functions.
For industrial automation engineers and system integrators, the design illustrates more than a memory-saving technique. It shows how hardware partitioning, bootloader changes and software-size optimization can work together to reduce external components in applications such as communication modules and connected drives.
Why consider a DDR-less EtherCAT slave?
EtherCAT is a real-time industrial Ethernet technology in which the master sends frames through the network. Slave devices process their relevant data as the frame passes, reading or inserting information along the way. This on-the-fly approach supports deterministic communication for devices including I/O modules, sensors, PLCs and drives.
In a conventional slave architecture, an EtherCAT ASIC or FPGA may be paired with an external processor for application-level work. That arrangement offers flexibility, but adds components and board requirements. The AMIC110 reference design combines industrial communication functions in a multiprotocol SoC and omits both an external EtherCAT ASIC and external DDR memory for this implementation.
Removing external memory also avoids the transfers associated with accessing it. The design describes this as a way to reduce memory-access latency and improve system performance, particularly for applications where EtherCAT communication and application processing share a compact hardware platform. Actual design choices still depend on the memory and processing needs of the end equipment.
How the AMIC110 divides EtherCAT work
The AMIC110 includes an Arm Cortex-A8 processor and a separate PRU-ICSS. In the reference design, the PRU real-time cores perform EtherCAT data-link tasks, including datagram processing, address mapping, error handling, distributed-clock functions and host interfacing. They also emulate the EtherCAT register space in internal shared memory.
The Arm processor runs the EtherCAT slave stack and the application layer, whose functions depend on the target device. The physical layer connects to the Ethernet network through DP83822 PHY devices. Together, these layers separate network signaling, deterministic frame handling and end-equipment application processing.
This division is useful when evaluating a communication architecture: tasks with strict timing requirements can be assigned to the PRUs, while application logic runs on the Arm core. The AMIC110’s programmable industrial communication subsystem also supports multiple industrial Ethernet and fieldbus protocols, although a specific implementation must be configured for its intended protocol and application.
Protocol and system capabilities
The reference design specifies two MII ports between the PHY and the AMIC110 MAC, up to eight Fieldbus Memory Management Units (FMMUs) and Sync Managers, and 8 KB of process-data RAM from PRU shared memory. Distributed-clock support includes SYNC0, SYNC1, LATCH0 and LATCH1 signals. The design reports support for the listed EtherCAT commands and passing the EtherCAT Technology Group’s Conformance Test Tool.
These details help define the scope of the reference implementation. They are not a substitute for checking the requirements of a target product, including its process-data needs, application workload, network topology and conformance obligations. The design identifies potential application areas such as robot communication modules, PLC-related communication, and wired or wireless AC and servo drive communication.
Fitting the software into on-chip memory
Operating without DDR requires coordinated optimization across the boot process, operating system, drivers and EtherCAT application. The reference design’s available on-chip memory is limited, so the software image must be reduced and placed carefully.
Trim the bootloader and manage PRU firmware
The second-stage bootloader is modified to remove functions that are unnecessary for this fixed DDR-less configuration. The described changes include bypassing DDR setup, removing board and SoC detection, and eliminating console utilities so the application can boot directly. The bootloader also wakes the PRUs and loads their firmware into PRU instruction RAM, avoiding the need to retain firmware-copy arrays in on-chip memory during application operation.
EtherCAT EEPROM data and PRU firmware binaries are stored in SPI flash and loaded as part of startup. The application’s memory sections can then reuse space occupied by the bootloader after the bootloader has completed its work. This is an example of using the lifecycle of each memory region—not only its total size—to make a constrained memory map viable.
Reduce application and runtime overhead
The software is further compacted by removing unused device, board and RTOS features. In the documented configuration, examples include eliminating unused I2C, SPI write and UART or text-console support, along with reducing logging, exception handling and stack-checking overhead where those functions are not required for the runtime application.
Build settings also matter. The design uses Thumb mode and size-oriented compiler optimization, removes debugging information, enables link-time optimization, and places functions and data into separate sections to support more deliberate placement. Stack and heap reservations are sized according to the application’s needs. These measures require careful validation: removing a feature or reducing a memory allocation is appropriate only when the product’s actual runtime and diagnostic requirements allow it.
What engineering teams should take away
The AMIC110 implementation provides a practical architectural example for compact industrial communication hardware: assign deterministic EtherCAT data-link processing to PRU-ICSS, keep application and protocol-stack responsibilities distinct, and optimize startup and software components together to fit internal memory. The reported conformance result shows that the specific reference implementation passed the ETG CTT; product teams must still validate their own configuration and application.
For design reviews, the key question is not simply whether external DDR can be removed. Teams should assess the complete memory budget, protocol features, application workload, boot flow, diagnostics and interface requirements. When these constraints align, a DDR-less EtherCAT slave architecture can reduce external component needs while providing a clearly partitioned path for real-time communication and application processing.