An optimized secondary ion mass spectrometry (SIMS) technique was used to characterize the solid source diffusion of dopants, i.e., arsenic and boron, from agglomerating cobalt disilicide sources. It was found that interface roughness plays a dominant role in determining the dopant profiles observed by SIMS. Double convolution of a step function with an exponential function, which represents the SIMS cascade mixing, and then again with a Gaussian function, which represents the silicide interface roughening, were performed to fit the SIMS silicon signal in the transition region of the silicide/silicon interface. Experimental profiles were then deconvoluted with the same functions to derive the ‘‘true’’ dopant diffusion. This convolution methodology eliminates the broadening factors provided by the measuring instrument and by the sample itself, and therefore results in a more physically meaningful diffusion profile. The broadening of SIMS silicon signal also provides a very precise characterization of the thermal stability of the thin polycrystalline silicide film. An analytical model was established to relate the SIMS silicon broadening and the sheet resistance of the agglomerated silicide film. This model predicts that while the amplitude of the root‐mean‐square roughening goes as the square root of time, the increase in sheet resistance varies linearly with time, provided that the amplitude of roughening is considerably smaller than the thickness of the silicide film.