Backup method for key data in hybrid SSD

[0024] Example

[0025] Mixed SSD with SLC-TLC as an example, see figure 1 , Is a read operation flow chart of the present embodiment, starting from step 101, then:

[0026] At step 102, after the read request arrives, the SSD will split a large read request into a plurality of small read sub-requests separately.

[0027] At step 103, the SSD controller is confirmed to determine if the requested data has a backup as the backup tag issued by the request. If the requested data is not backed up, step 104 is performed, and the read sub-request is processed according to the hybrid SSD, and then proceeds to step 109; if the requested data has a backup, step 105 is executed, from the SLC read data, Then go to step 106.

[0028] At step 106, it is judged whether or not read error occurs when the SLC read data. If a read error occurs, step 107 is performed from the TLC, and then perform step 108, perform data recovery with the backup data saved on the TLC, and finally proceeds to step 109; if no read error occurs, directly go to step 109.

[0029] Mixed SSD with SLC-TLC as an example, see figure 2 It is a write operation flowchart in the present invention, starting from step 201, then:

[0030] At step 202, after the request arrives, the SSD is split by a large write request into a plurality of small write sub-requests separately.

[0031] At step 203, the SSD controller checks if the data issued by the write request is determined whether the requested data requires a backup. If the requested data does not require a backup, step 204 is performed, and the write sub-request is processed according to the hybrid SSD, and then proceeds to step 210; if the requested data requires a backup, step 205 is performed.

[0032] At step 205, it is judge whether or not the size of the request belonging the current sub-request is greater than the threshold. If it is greater than the threshold, then the request does not need to be accumulated, performing step 206, directly writing each of the data to the SLC and TLC; if less than or equal to the threshold, the request needs to be cumulative backup, proceed to step 207.

[0033] At step 207, the data of the sub-request is written on the SLC and adds the cumulative page number ACC_COUNT 1, and then performs the next step 208.

[0034] At step 208, it is judged whether the accumulated page number ACC_CONT reaches the number of TLC strips. If so, step 209 is performed, one of the cumulative data is migrated to the TLC, and the ACC_CONT is set to 0, and the proceeds to step 210 end; if it is not yet reached, the direct proceeds to step 210.

[0035] image 3 It is a schematic diagram of a TLC strip in the present invention.

[0036] Here you need further explanation of the TLC strip. Since there is a channel, a chip, a wafer, a group 4 parallel in the SSD, the different sub requests of the same request will be assigned to different parallel units on parallel. For TLC, since the physical page size is 16KB, each 4 logical pages correspond to a physical page, when the request is assigned to a different channel, chip, wafer, and group, if the request size is requested Do not align, it is very likely that some physical pages are not brought to 4 logical pages, resulting in large writing among TLCs. like image 3As shown, when a request containing 16 logical pages is written to TLC, these logical pages correspond to two channels, 2 chips, 4 wafers, 8 packets of 8 physical pages, and each physics The pages will only have two logical pages, so there is a double write amplifier. In the present invention, the TLC strip size is the minimum amount of data across all parallel units of the TLC, and the number of TLC strips can be written to a 4kb page number of a TLC strip, which is set to the threshold of the migration during the accumulation process. Migration with TLC strip size can completely avoid the writing of TLCs caused by small request backups, achieving ideals to reduce the effect of write an enlargement. The TLC strip size is determined by the parallel architecture and allocation strategy of SSD. exist image 3 In the architecture shown, the TLC is distributed on 2 channels. Each channel has a TLC chip, each chip has 2 wafers, each wafer has two 2 packets, two red dashed lines The content is combined for a TLC strip, a total of 2 * 1 * 2 * 2 = 8 16KB pages, that is, 32 4KB pages, the TLC strip page number is 32, the TLC strip size is 128KB. That is, when the number of pages accumulated in the SLC reaches 32, the accumulated data is migrated to the TLC, and the TLC is not written in this process.