We studied the feasibility of an experiment searching for sub-millicharged particles ($chi$s) using 30 GeV proton fixed-target collisions at J-PARC. The detector is composed of two layers of stacked scintillator bars and PMTs and is proposed to be installed 280 m from the target. The main background is a random coincidence between two layers due to dark counts in PMTs, which can be reduced to a negligible level using the timing of the proton beam. With $N_textrm{POT}=10^{22}$ which corresponds to running the experiment for three years, the experiment provides sensitivity to $chi$s with the charge down to $5times10^{-5}$ in $m_chi<0.2$ $textrm{GeV}/textrm{c}^2$ and $8times10^{-4}$ in $m_chi<1.6$ $textrm{GeV}/textrm{c}^2$. This is the regime largely uncovered by the previous experiments. We also explored a few detector designs to achieve an optimal sensitivity to $chi$s. The photoelectron yield is the main driver, but the sensitivity does not have a strong dependence on the detector configuration in the sub-millicharge regime.