We have developed a solution-grown bulk SiC crystal growth method [1]. This method can grow SiC crystals of high quality because the process is close to thermal equilibrium. Moreover, by forming macrosteps on the surface, the threading dislocations in the crystal are converted into basal plane defects, and these defects are ejected out of the crystal during the growth process, thereby reducing the dislocation density [2-4]. If macrosteps are developed too much, they will conversely cause macrodefects such as inclusions [5]. In other words, it is important to maintain a moderate macrostep height. In controlling the macrostep height, the flow distribution in the solution is important [6]. In order to achieve a proper flow distribution over the crystal growing surface, it is necessary to intermittently change the crystal growing conditions to periodically realize multiple flow states on the crystal growing surface. We call this method “switching flow method”. However, there are so many crystal growth parameters that it is almost impossible to optimize the combination of multiple sequences by hand. For such complex control, we have used machine learning techniques to build a digital twin of the actual crystal growth apparatus in a computer and repeatedly run virtual experiments with it to optimize the parameters [7] In this study, we used this technique to optimize the growth conditions and attempted to grow 8-inch crystals. Top-Seeded Solution Growth (TSSG) method used in this growth. In this method, crystals are grown in a Si-based solvent in a carbon crucible at high temperature. A temperature distribution and a flow distribution are controlled by the position and rotation of the crucible and seed crystal, the power of the RF power supply, and so on. The experimental conditions for 8-inch crystal growth were optimized as follows: (1) The temperature and flow distributions under the optimum good condition of 6-inch crystal growth we have achieved are calculated by simulation. (2) The distributions of the optimized 6-inch growth are represented in latent space by the Variational Auto-Encoder (VAE) method. (3) Using an 8-inch setup, simulations are performed under various crystal growth conditions, and the simulations are represented in latent space in the same way. (4) The conditions for 8-inch growth that can obtain the same distribution as the optimum 6-inch condition are sought by changing the conditions in latent space. [7] The seed crystals were 8-inch crystals turned off once at [11-20]. Si-Cr solvent was used as the solvent.