*3.2. Analysis of Prediction Model*

## 3.2.1. Prediction Object Selection

The object of programming operation is flash memory page, and scholars have focused more on prediction research in the past. However, various disturbances seriously affect the accuracy of these prediction models in actual scenarios. This paper takes the flash memory block as an independent prediction object. The decision was based on the following reasons:

#### 1. Obvious Endurance Difference between Flash Memory Pages

As shown in Figure 2a, take Intel's 29F32B2ALCMG3 NAND Flash particles as an example, different pages show different RBE numbers. In the interval where the page number is lower than 200, the RBE numbers of some pages is significantly higher, showing a more obvious trend of rising with the increase in number of P-E cycles. While other flash memory pages have significantly lower RBE numbers, the change trend is also very irregular. Even after ignoring the high-frequency jitter, there is a local feature where the RBE numbers decreases with the increase in the number of P-E cycles.

**Figure 2.** The relationship between the RBE on the page and the P-E cycle number: (**a**) Early and mid in lifecycle; (**b**) End of lifecycle.

The memory cells of different pages have differences in structural size and attributes [19], which leads to inconsistency of the actual tunneling charges suffered by different pages under the same macro programming pressure, directly affecting the degree of cell wear. Moreover, the new three-dimensional multilayer process will bring more serious physical structure differences and disturbance effects [20], resulting in more significant differences in the endurance of flash memory pages. If the flash memory page is used as the prediction object for modeling, the difference in the degree of wear and change trend will greatly reduce the accuracy of the model's prediction results.
