Experimental techniques

Experiments were carried out using a customized commercial LT-STM/AFM system (Omicron) operating at 4.5 K and controlled by Nanonis electronics. We used highly arsenic-doped silicon (100) samples (Virginia Semiconductor, Inc.) with a resistivity of 3–4 mΩ cm (∼1.5 × 10⁻¹⁹ atom cm⁻³). The H–Si(100) samples were prepared in a ultrahigh vacuum chamber with a base pressure of 3 × 10⁻¹¹ Torr. They were first degassed for about 12 h at ∼600 °C then flashed at high temperature (1,250 °C) before hydrogen termination. The followed sample preparation procedure, described in detail in ref. ⁴⁶, ensured a high-quality H–Si(100) surface with low-defect density.

AFM data were acquired using a qPlus sensor equipped with a separate wire for tunnelling current^11,47, showing a Q factor ∼10,000 and a resonance frequency of 26 kHz. Sharp tips were obtained by electrochemical DC etching in NaOH solution the 50-μm-thick polycrystalline tungsten wire glued to the sensor. The qPlus was cleaned in ultrahigh vacuum in a FIM attached to our LT-AFM system. It first underwent a series of ebeam heatings with the tip at 500 V and 1 mA for about 15 s, then was further cleaned by field evaporation in FIM²⁹.

The AFM images presented in this paper were acquired in constant height mode, that is, frequency shift maps, with an oscillation amplitude of 1 Å. Care was taken to minimize drift during the image acquisition, typically of ∼20 min per image, by settling for about 12 h after the approach to allow piezo stabilization, and using an atom tracking module as implemented in the Nanonis controller. The sensor's oscillation amplitude was calibrated using the tunnel current method⁴⁸. To avoid cross-talk problems and artifacts in frequency shift measurements, for example, phantom force, all AFM data were acquired at 0 V (refs ^{47, 49}).