Data retention (a time-variant characteristic of 3-D- NAND flash memory) is predicted through a bi-directional long short-term memory (LSTM) neural network (NN) model that learns sequential data obtained from chip measurements of a triple-level cell (TLC). The predicted results for all time points of each program (PGM) state are accurately predicted by the threshold voltage ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${V}_{\text {th}}$ </tex-math></inline-formula> ) distribution. Thus, the predicted <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${V}_{\text {th}}$ </tex-math></inline-formula> can be used to analyze the cause of retention failure. When the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${V}_{\text {th}}$ </tex-math></inline-formula> of the target cell is high or when that of the adjacent cell is small, the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${V}_{\text {th}}$ </tex-math></inline-formula> loss of the target cell is large. In addition, the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${V}_{\text {th}}$ </tex-math></inline-formula> loss increases as the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${V}_{\text {th}}$ </tex-math></inline-formula> of the adjacent cell decreases. Using a fully calibrated TCAD simulation, we verify the NN-based <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${V}_{\text {th}}$ </tex-math></inline-formula> prediction by checking the change in the electron concentration in the nitride layer. Furthermore, the NN model predicts the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${V}_{\text {th}}$ </tex-math></inline-formula> for cells existing in other blocks, showing that they are consistent with the measured <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${V}_{\text {th}}$ </tex-math></inline-formula> . The prediction times were 5 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\times \,\,10^{{5}}$ </tex-math></inline-formula> s, 5 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\times \,\,10^{{6}}$ </tex-math></inline-formula> s, and 2 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\times \,\,10^{{6}}$ </tex-math></inline-formula> s, but using machine learning (ML), we reduced the time required to predict the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${V}_{\text {th}}$ </tex-math></inline-formula> to only 2 s. Therefore, the proposed ML method enables fast, accurate, and effective predictive modeling of the time-variant <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${V}_{\text {th}}$ </tex-math></inline-formula> of 3-D TLC NAND flash memory. Finally, the predicted <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${V}_{\text {th}}$ </tex-math></inline-formula> can be included in the read retry table or included in the lookup table of the compensation circuit in NAND solutions. This can save a significant amount of time that would otherwise be spent on actual long-term measurements.