First order gradient errors are canceled with no current source splitting by placing consecutive current sources symmetrically around the center of the array. Consecutive elements that correspond to small input amplitudes (mid-scale codes) make a smaller spatial jump than those correspond to larger signal amplitudes. Both linear and second order gradients are reduced by splitting each current cell into two and placing sub-elements symmetrically with respect to the center of the array to address the linear gradient effect. To address second order gradients, current element placement follows a pattern such that consecutive element pairs are chosen with one of the pair being placed with respect to the zero error contour of the second order gradient so as to have a positive error and the second of the pair being placed so as to have a negative error resulting in reduced second order error accumulation.