Existing subthreshold slope methods are shown to be far from accurate in extracting gate capacitive coupling coefficient alpha(G) in stacked gate flash memory cells. The origin of the error is systematically identified: (i) process variations induced mismatch and (ii) underlying bulk capacitive coupling. To alleviate such drawbacks, a new version of the subthreshold slope method at room temperature is established: alpha(G) = 0.06(n(f) - alpha(B))/s(f), where the subthreshold swing s(f) is from flash memory cells, the subthreshold slope factor nf is from dummy transistors via threshold voltage against source-to-substrate bias measurement, and the bulk coupling coefficient alpha(B) is from a linear extension of the dimensional dependencies in the literature. The resulting alpha(G) of around 0.55 again agrees consistently with those dependencies and once drain and source coupling experiment is performed, the relation of Sigma alpha(i) approximate to 1 is achieved for all involved coupling coefficients alpha(i)'s.. The sidewall source-side injection flash memory cells are also investigated. With the improved method, this manufacturing process is proved free of process variations issue and is characterized with alpha(G) of 0.374 and fringing capacitance of 0.204 fF. (C) 2004 Elsevier Ltd. All rights reserved.