What part of the CM-900 has overheated? Is it the big blocky voltage regulator between the large and small capacitors, the smaller voltage regulator in the corner by the two small capacitors, the STM32 between the big capacitor and the 2x5 header, the small tristate buffer, the RS-485 transceiver, etc.? The parts and layout are on pandora's OpenCM904 repository in the hardware folder.
Does the CM-900 respond at all (LEDs turn on/off when (un)plugging and/or pressing reset button), or is it only the dynamixel bus that is non-responsive?
A non-responsive dynamixel bus is caused by failure of the tri-state buffer used on
the CM-900 ES and CM-900 v1.x through static/over-voltage
damage (usually only the RX direction fails). There have been a couple people previously on the circle that had more significant hardware problems with the STM32 of the CM-900 ES and CM-900 v1.x. Not sure there are any solutions for those STM32 issues other than replacing the STM32 (and using a JTAG to upload the bootloader to the new STM32), but the tri-state buffer can be replaced with a proper hot-air rework station and the correct spare part. The two CM-900 v1.01 from the lab that have dead tri-state buffers produce no additional heat, but do not have the ability to read the dynamixel bus.