Electronic Devices Elena N. Akimova, Vladimir E. Misilov, and Maxim S. Arguchinsky 13 2 2024 https://doi.org/10.6025/ed/2024/13/2/49-56 https://www.dline.info/ed/fulltext/v13n2/edv13n2_2.pdf A memory-efficient algorithm is developed and implemented to address the reverse issue of determining the variable density within a curvilinear layer from gravitational information. This algorithm is structured around the stabilized bi-conjugated gradient approach. By approximating the SLAE matrix with a Toeplitz-block-Toeplitz structure, the modification dramatically lowers the need for memory. The parallel versions of the algorithm were created for the Uran supercomputer, utilizing a combination of MPI+OpenMP technology. The practicality of this algorithm was demonstrated through its successful application to a model problem featuring synthetic data.