For IBMQ device based on superconducting technology, when we transpile the circuit, it is decomposed of U1,U2,U3 and CNOT gates, which are its native gates. However for trapped ion device, its native gates are rx,ry,rxx and ms gate. Why do the 2 technologies choose different native gates?
The choice of gates is entirely dependent on the types of interactions that occur in the different architectures. The cross resonance gate used by IBM generates the ZX interaction you want (plus other stuff) that leads to a CNOT. Trapped ions have XX type interactions that give rise to Molmer Sorensen gates. For single qubit gates it depends on what driving terms you have available to rotate your qubits on the Bloch sphere. The U gates on IBM devices are just combinations of X(pi/2) pulses and Z rotations that get done in software. So the only drive channel one needs is an X drive.