A pulse generator project originally built for internal testing has been open-sourced directly: an FPGA plus STM32 dual-core design, up to 32 outputs, and a theoretical timing precision of 500ps. It does not take the high-speed transceiver route; instead it uses SelectIO plus parallel comparators to synthesize pulses, offering a different implementation path for multi-channel synchronized triggering.
RSPLG_101 uses an FPGA plus STM32 architecture, with the FPGA handling high-precision timing and pulse generation and the STM32 handling control and configuration. The device supports up to 32 pulse outputs; each channel can be configured independently or output synchronously; frequency, pulse width, delay, and polarity are all programmable. The project uses a 4-layer PCB with components placed on one side, which the author hopes will reduce hardware fabrication cost. The official theoretical frequency range is 0.5Hz to 50MHz, with a note that actual performance is limited by the IOs. 500ps is the theoretical timing precision given by the project; the author explicitly warns that actual pulse width and timing are also affected by FPGA IO, routing, port matching, and measurement method, and that the same level cannot be achieved under all output conditions.
At the core is a high-precision timer. Rather than using GT high-speed transceivers to chase higher frequencies, the author takes the SelectIO route, synthesizing pulses through multiple parallel comparators and then outputting them via OSERDES. The design uses 8 sub-timers, making the pulse output speed 8 times that of a single sub-timer, which is the key to achieving high-precision, multi-channel output. The cost is resource and routing pressure: the author notes that a single channel requires about 1000 CLBs, but routing occupancy is large, so it is recommended that 35T devices prioritize 16 channels and 100T devices consider 32 channels. The timer defaults to a 32-bit width and can be changed to 24 bits to reduce resource consumption when precision requirements are not high.
To improve timing precision, the project provides auxiliary signal clock calibration: clock sources include an external input clock, the STM32 PLL2 output clock, and the MS5351 output clock; usually the STM32 PLL clock is the main input, and it can also be switched by command. Users can measure the RF3 SMA interface signal with an oscilloscope and then fine-tune the clock frequency point via the rsplg.osctrim command. SPI is used between the FPGA and STM32, with a memory-mapped-like approach to accessing FPGA-side resources; parameter changes are completed by the control side, while pulse generation remains inside the FPGA hardware. Control methods include a serial command line, SCPI commands, and a companion host application. The source code provides HDL and C; the complete Vivado project is not given directly due to upload limits, and build scripts are provided instead; the original project used Vivado 2020.2, Ubuntu 20.04, a Xilinx Artix-7 7A100T device, and a Windows 10 plus AC6 compiler environment for the STM32 software.
For domestic engineers working on multi-channel synchronized triggering, the value of this project lies in its implementation approach rather than direct productization: the SelectIO plus parallel comparator structure can bypass high-speed transceivers and perform multi-channel pulse synthesis on devices such as Artix-7. When reproducing it, note several hard constraints: the FPGA Flash must use the corresponding 128Mbit model and cannot be directly replaced with 256Mbit; the STM32 BOOT setting must be matched; there is no pull-down resistor at the output, so when connecting to a device under test, pull-down and ringing must be handled separately. 32 channels place great pressure on synthesis and place-and-route, and changing the device model or development environment requires re-adaptation. It does not involve supply chain substitution, but the HDL and C source code can serve as a reference starting point for self-developed timing logic.