Machine learning-assisted design of zero magnetic temperature-control chips for on-chip quantum sensors

Alkali-vapor atom-based quantum sensors are urgently needed in the cutting-edge frontier of extremely weak field measurements. However, the generation and control of atomic vapor currently heavily rely on the flexible heater and the discrete temperature sensor, which severely hinder practical application. Herein, we report on-chip, high-performance temperature-control chips with very low residual magnetic flux density, enabled by the multipole-field suppression microstructure designed by particle swarm optimization algorithms, achieving two to three orders of magnitude reduction compared with conventional structures. The unique structure is formed by multiturn parallel winding arrangement, resulting in a synergistic cancelation of the stray magnetic field. A one-step manufacturing process based on oxygen plasma treatment and metal lift-off facilitated tough adhesion is developed for high-efficiency fabrication. As a result, a low magnetic field coefficient of ∼4.3 pT/mA, a high heating power of ∼1 W, and an exceptional temperature fluctuation of ±10 mK are reached, which represents significant improvement in comprehensive performance compared to previous reports. Furthermore, we configure a chip-scale quantum magnetometer with the optimal sensitivity of 10 fT/Hz1/2. Benefiting from its high-performance capabilities, integrated functionality, and the straightforward microfabrication process, the methods reported here are, in principle, generalizable to all thermal alkali-atom-based quantum sensors, including atomic magnetometers, atomic clocks, atomic gyroscopes, and beyond.