The photoactive layer of polymer solar cells is commonly processed from a four-component solution, containing a semiconducting polymer and a fullerene derivative dissolved in a solvent cosolvent mixture. The nanoscale dimensions of the polymer fullerene morphology that is formed upon drying determines the solar cell performance, but the fundamental processes that govern the size of the phase-separated polymer and fullerene domains are poorly understood. Here, we investigate morphology formation of an alternating copolymer of diketopyrrolopyrrole and a thiophene-phenyl-thiophene oligomer (PDPPTPT) with relatively long 2-decyltetradecyl (DT) side chains blended with [6,6]-phenyl-C-71-butyric acid methyl ester. During solvent evaporation the polymer crystallizes into a fibrous network. The typical width of these fibers is analyzed by quantification of transmission electron microscopic images, and is mainly determined by the solubility of the polymer in the cosolvent and the molecular weight of the polymer. A higher molecular weight corresponds to a lower solubility and film processing results in a smaller fiber width. Surprisingly, the fiber width is not related to the drying rate or the amount of cosolvent. We have made solar cells with fiber widths ranging from 28 to 68 nm and found an inverse relation between fiber width and photocurrent. Finally, by mixing two cosolvents, we develop a ternary solvent system to tune the fiber width. We propose a model based on nucleation-and-growth which can explain these measurements. Our results show that the width of the semicrystalline polymer fibers is not the result of a frozen dynamical state, but determined by the nucleation induced by the polymer solubility.