Aetina Ships Jetson Thor Boxes With an EtherCAT Master Port
Putting an EtherCAT master on the front panel of a 2,070 TFLOPS inference box collapses the usual two-controller split where a Jetson thinks and a separate PLC moves.
Aetina announced general availability of the AIE-KT78 and AIE-KT68 on 17 September, edge AI systems built on NVIDIA Jetson T5000 and T4000 modules. The KT78 carries the T5000 at 2,070 FP4 TFLOPS with 128 GB of LPDDR5X; the KT68 carries the T4000 at 1,200 FP4 TFLOPS with 64 GB.
Most of the specification is what you would expect from an inference box: QSFP28 at 4x25 Gbps, dual 10GbE on RJ45, up to eight GMSL2 camera inputs, USB 3.2, isolated digital I/O and M.2 expansion. Power is 9 to 48 VDC, operating range -25 to +55 degrees Celsius, chassis thickness 80 mm.
The part worth noticing is a dedicated 1GbE EtherCAT master port, with microsecond-level synchronisation to motors, joints, sensors and actuators. That is not a detail a datacentre product would carry, and it is the reason this box is interesting to anyone building machines.
The common architecture in Jetson-based robotics has been two controllers: the Jetson runs perception and planning, a separate PLC or motion controller runs the axes, and the two exchange messages across a boundary that adds latency and a synchronisation problem nobody enjoys debugging. An EtherCAT master alongside the inference engine, in one -25 to +55 C enclosure, removes that split — perception, decision and motion on the same clock.
Richard Hung, vice president of Aetina's product division, described the systems as turning Jetson Thor's compute into a compact, mass-production-ready system. Target applications are collaborative robots, humanoid robots and autonomous machines, and the units are shipping now.