Superjunction (SJ) power devices have shown to be more efficient than their planar counterparts, with 600 V Si SJ-MOSFETs showing a 20% lower specific on-resistance (Ron,sp) than equivalent planar MOSFETs.[1] A superjunction is a drift region structure comprised of alternating p-doped and n-doped columns that charge-balance in the depleted state.[2] This permits higher doping in the epilayer, reducing Ron,sp for equivalent thickness devices. Since SiC is capable of improving Ron,sp by a further 10-fold due to its higher critical electric field, SiC superjunction devices could potentially have advantages in the high-voltage (>1.7 kV) range. However, current SJ fabrication methods are designed for Si and are practically inapplicable to 4H-SiC device fabrication.[3] Trench filling epitaxy (TFE) is where deep trenches are etched into n-doped (p-doped) SiC and epitaxially refilled with p-doped (n-doped) SiC, and is a feasible approach to SiC-SJ fabrication even for a drift region up to 50 μm thick. However, further development is needed for TFE to be used in practice. One challenge of TFE is the occurrence of pattern and etch imperfections prior to epitaxy, such as microtrenches, surface roughness or trench/ mesa width disparity. Annealing trenches under various atmospheres can round mesas and smooth surfaces, making it a useful trench conditioning method to alleviate inevitable patterning flaws.[4,5] Here, we present a systematic study of annealing 4H-SiC trenches in Ar, H2 and HCl at varied temperature and duration, to derive structure-condition relationships that will drive TFE process design. Focus is placed on H2 and HCl due to their relevance in chemical vapor deposition (CVD) as this enables trench modification in situ as a preconditioning step, prior to CVD epitaxy. In this work, trenches were processed on the Si face of n+ 4H-SiC (0001) substrates off-cut by 4° in the [11-20] direction. These were masked using 500 nm SiO2 with TEOS as a precursor by low pressure (LP-) CVD, then a sputtered Ni layer was patterned by photolithography. Using a F-based ICP etch in an Oxford Instruments PlasmaPro 100 Cobra etcher, trenches of ~5 μm depth were produced in the SiC [11-20] substrate directions using the SiC/SiO2/Ni mask. Resulting trenches were as low as 1.5 μm wide with corresponding mesas of 2.5 μm, which is termed a 4 μm pitch. Pitches of 8 μm and 20 μm pitch were also investigated. Annealing was performed in an LPE ACiS M8 RP-CVD reactor between 1550°C and 1650°C for up to 4 hours. By comparing scanning electron micrographs (SEM) of trench cross-sections cleaved along the [1-100] direction, the effect of annealing on trench profile can be examined. At 1550°C, Ar (30 slm) and HCl (0.1 slm, with H2 100 slm carrier flow) have no effect on the trench shape but H2 (100 slm) causes a duration-dependent faceting of the mesa corners (see Fig. 1). This appears as a ‘truncation’ of corners at 30-60° from the (0001) surface along the [1-100] direction. This angled facet extends in length at a rate of ~0.3 μm h-1 at the mesa top but only by ~0.1 μm h-1 at the trench bottom, which may be caused a differential etching rate of SiC by H2 depending on the trench depth or sidewall slope. This faceting is also observed in trenches refilled by CVD at 1550°C with H2 carrier gas, which we propose will encourage fusing of the epilayer over trenches to produce a flat end surface required for planarization and further processing. Corner faceting at the trench bottom can also remove microtrenches after 10 min for trenches of 4 μm pitch (Fig. 1). In addition to faceting, H2 also anisotropically etches mesas depending on the trench depth and the trench direction relative to the 4H-SiC crystal orientation. See image 2b), “wheel” structures were also on the same back, so that any effect of crystallographic direction could be investigated. For trenches aligned to the [11-20] direction, mesa tops are laterally etched ~0.2 μm h-1 faster than at the trench bottom, which may be due to a microloading effect, whereby the concentration of H2 gas is lower at the trench bottom. Similarly, the trench depth decreases by ~0.3 μm h-1, indicating that mesa tops are vertically etched faster than trench bottoms and that H2 etching is directionally isotropic at a given depth. Examination of the lateral etch rates for trenches aligned 0-180° to [11-20] shows that the rate is slowest in the 〈11-20〉 (a-plane) and 〈1-100〉 (m-plane) directions, compared with intermediate angles (Fig. 2). This is an important consideration for both process and device design. From atomic force microscopy (AFM), it is found that the RMS surface roughness is <40 nm for all H2 annealed samples after 10 min, indicating smoothing is fast and does not further improve with annealing duration at 1550°C. In the full submission, we report all of the effects found with the various gases and temperatures. This is then combined with TFE to highlight the potential uses of the pre-TFE treatment as an aid to complete refill of the trenches, enabling SiC superjunction technology.