Multi‐level non‐volatile ferroelectric memories are emerging as promising candidates for data storage and neuromorphic computing applications, due to the enhancement of storage density and the reduction of energy and space consumption. Traditional multi‐level operations are achieved by utilizing intermediary polarization states, which exhibit an unpredictable ferroelectric domain switching nature, leading to unstable multi‐level memory. In this study, a unique approach of composition‐graded ferroelectric ScAlN to achieve tunable operating voltage in a wide range and attain precise control of domain switching and stable multi‐level memory is proposed. This non‐volatile memory supports multi‐level storage up to 7‐bit capacities, and exhibits enhanced performance compared to the uniform composition device, showing one order of magnitude higher ON/OFF ratio, 30% reduced working voltage, and up to 50% enhanced tuning window of operating voltage. Finally, the emulation of long‐term plasticity and linear weight update akin to biological synapse with high uniformity and reliability are demonstrated. The proposed composition‐grading architecture offers new opportunities for next‐generation multi‐level ferroelectric memories, paving the way for advanced hybrid integration in multifunctional computing systems.