This article explains how bad blocks are generated, how SSDs detect and manage bad blocks, what problems exist in the bad block management policy recommended by vendors, which management methods are better, how formatting hard disks do not cause bad block table loss, and what security risks are generated after SSD repair.
Bad block management is related to SSD reliability and efficiency. The bad block management practices provided by Nand Flash manufacturers may not be very reasonable. In product design, if some abnormal conditions are not considered properly, some unexpected bad blocks will often occur.
For example, after testing many different master SSD found the new bad block due to abnormal power problem is very common, using the search engines search “abnormal power produce bad block” or similar words, will find that the problem exists not only in the testing process, the actual problems happened in the end-user is very much also.
Who manages the bad block?
For a controller without a dedicated flash file system, bad blocks can be managed by SSD controller firmware, or for a dedicated flash file system, bad blocks can be managed by a dedicated flash file system or Driver.
Three types of Bad blocks
1, Factory bad block, or initial bad block, that is, the factory because it does not meet the manufacturer’s standards or the manufacturer’s random testing over the block can not meet the manufacturer’s published standards, in the factory has been identified by the manufacturer as a bad block; some factory bad block can be Erase, some can not be Erase.
2, The use of the process because of wear and tear caused by the addition of bad blocks or the use of bad blocks.
3, False bad blocks misjudged by the master for reasons such as abnormal power loss.
Not all new bad blocks are caused by wear. If the SSD does not have the abnormal power failure protection function, the main control may misjudge the bad blocks or generate new bad blocks. In the case of no abnormal power failure protection, if the Lower page has been successfully programmed and the Upper page is suddenly powered off, the data on the Lower page will be sent incorrectly. If the number of data errors exceeds the ERROR correction capability of SSD ECC, the data will be read incorrectly. Blocks are identified as Bad blocks by the master and added to the Bad Block table.
Some new bad blocks can be erased. After a new bad block is erased, data erasers may not fail again because the error depends on the pattern in which data is written. If a pattern fails, another pattern may fail.
Percentage of factory bad blocks in the total Device
We have consulted several Nand Flash manufacturers about the ratio of factory bad blocks in the whole Device, and a common statement is given: The ex-factory bad block ratio is not more than 2%, and the manufacturer will leave a part of the allowance to ensure that even after reaching the maximum P/E times promised by the manufacturer, there is still a bad block ratio of no more than 2%. It seems that it is not easy to guarantee 2%. When receiving a new sample, the bad block ratio is more than 2%, and the actual test is 2.55%.
The bad block determination method
Judgment method of factory bad block
Bad block scanning basically checks whether the byte corresponding to the address specified by the manufacturer has the FFh flag. If the byte does not have FFh, the bad block is detected.
The location of the bad block identifier is roughly the same for each vendor, but different for SLC and MLC. Take Micron as an example:
1.1 For SLC with small pages (528Byte), whether the sixth Byte in the spare area of the first page of each Block has the FFh flag, and if not, it is a bad block. 1.2 For SLC with large pages (greater than or equal to 2112Byte), whether the first and sixth Byte of the spare area of the first page of each Block has the FFh flag, and if it does not, it is a bad block. 1.3 For MLC, factory bad blocks are scanned by scanning whether the first or second Byte of the spare area of the first page and last page of each Block has the 0xFF flag, and if it is 0xFF, it is good and fast, and if it is not 0xFF, it is a bad block.
To illustrate, borrow a diagram from Hynix Datasheet:
What data is in the bad block? All zeros or all ones? Of course, this may not be true. The factory bad block may be true, but not the new bad block. Otherwise, it is impossible to hide data through a “bad block” :
Can the factory bad blocks be erased?
Some are “allowed” to erase, while others are prohibited by the vendor. “Allowed” erasure only means that the bad block identifier can be changed by sending an erasure command, rather than suggesting that the bad block can be used.
Vendors strongly recommend that you do not erase bad blocks. Once the bad block flag is erased, it cannot be “recovered”. It is risky to write data on the bad block.
In the process of using it, added the bad block determination method
A new bad Block is added to determine whether the operation on Nand Flash is successful based on the feedback result of the status register. When Program or Erase is reported as FAIL, the SSD master will classify the Block as a bad Block.
2.1. An error when executing the erase command.
2.2. An error when executing the write command.
2.3、Error when executing read command; when executing read command, if the error number of bit exceeds the error correction capability of ECC, the Block will be judged as a bad block.
Bad block management methods
Bad blocks are managed by creating and updating the Bad Block Table (BBT). Some engineers use one table to manage the factory bad block and the newly added bad block. Some engineers manage the two tables separately. Some engineers use the original bad block as a separate table and use the factory bad block and the newly added bad block as another table.
For the contents of the bad block table, the expression is not consistent, some of the expression is rather rough, for example, use 0 to indicate good fast, use 1 to indicate bad block or vice versa. Some engineers use more detailed descriptions, such as 00 for factory bad blocks, 01 for bad blocks when Program fails, 10 for bad blocks when Read fails, and 11 for bad blocks when Erase fails.
Bad block tables are typically saved in separate areas (e.g. Block0, page0 and Block1, page1), each time after the power directly reads BBT, efficiency is relatively high, considering the Nand Flash itself will also be damaged, which may lead to the loss of BBT, therefore, usually will BBT backup processing, backup how many copies each is different, some backup 2 copies, also some backup 8 copies, In general, you can use the voting system of probability theory to do the calculation, but at least two.
Bad block management policies include bad block skipping policies and bad block replacement policies.
Bad block skipping policy
- For the initial bad block, the bad block skip will skip the corresponding bad block through BBT and directly save the data to the next good block.
- Update the new bad block to BBT and transfer the valid data in the bad block to the next good block. Skip the bad block every time you perform Read, Program, or Erse.
Bad block replacement strategy (recommended by one Nand Flash vendor)
Bad block replacement refers to using good blocks in the reserved area to replace bad blocks generated during use. If an error occurs on the NTH page in the program, then under the bad Block replacement policy, the data from page0 to Page (n-1) will be copied to the same position in the reserved free Block (such as Block D). Then write the NTH page in the data register to Page N in Block D.
Vendors recommend that the entire data area be divided into two parts: the user-visible area for normal data operations, and the spare area for bad block replacement for storing the bad block replacement data and storing the bad block table. The spare area accounts for 2% of the total capacity.
When Bad Block has an FTL will Bad Block address remap to reservation Block address, rather than just skip the Bad Block to the next Block, every time before the logical address writes operation, will first calculate what physical address can be written, which address is Bad Block, if it is a Bad Block, put the address to write data into the corresponding resources?
DiskMFR did not see suggestions on whether the 2% reserved area should be included in the OP area or additional areas, nor did he see a description of whether the 2% reserved area is dynamic or static. Adding the 2% reserved area is an independent area and a static area, so this approach will have the following disadvantages:
- Directly reserving 2% of the area for bad block replacement will reduce the available capacity and waste space. At the same time, due to the small number of available blocks, the average wear times of usable bad are accelerated;
- If the available area bad blocks exceed 2%, that is, all reserved areas are replaced, and the generated bad blocks cannot be processed, the SSD ends its life.
Bad Block Replacement Policy (some SSD vendors)
In fact, in real product design, it is rarely seen that 2% of the bad block is reserved as the bad block replacement area. Generally, free blocks in the OP (Over Provision) area are used to replace the bad block added during the use process. Take garbage collection for example. Erase the Block. If the Erase status register fails to Erase the Block, the bad Block management mechanism updates the address of the Block to the list of new bad blocks. Write the valid data pages in the bad Block to the Free Block of the OP area, update the bad Block management table, and skip the bad Block to the next available Block when writing data next time.
OP different vendors to the size of the set are different, different application scenarios, for different reliability requirements, OP size will be different, there is a trade-off relationship between OP and stability, the greater the OP, in the process of writing and garbage collection available space, the greater the performance is more stable, shows the performance curves of the smooth, on the contrary, the smaller the OP, performance stability is poorer, Of course, more space is available to the user, and more space means lower costs.
Generally speaking, OP can be set at 5%-50%, and 7% OP is a common proportion. Different from the 2% fixed Block suggested by the manufacturer, 7% OP is not fixed in some blocks, but dynamically distributed in all blocks, which is more conducive to the wear balance strategy.
SSD maintenance problems
For most SSD manufacturers that do not have the master control technology, if the product is repaired, the common practice is to replace the faulty device and mass produce it again. In this case, the list of newly added bad blocks will be lost. Losing the newly added bad block table means that the Nand Flash that has not been replaced already has bad blocks. The operating system or sensitive data may be written into the bad block area, which may cause the user’s operating system to crash. Even for a vendor with a controller, it depends on the vendor’s attitude towards the user to save the list of existing bad blocks.
Whether bad block generation affects the read/write speed and stability of SSDS
Factory bad blocks are independent of the bitline, so the erasability speed of other blocks is not affected. However, if enough new bad blocks are added to the SSD, the number of available blocks in the SSD decreases, which increases the number of garbage collection times. Meanwhile, the OP capacity decreases, which seriously affects the garbage collection efficiency. When bad blocks increase to a certain level, SSD performance deteriorates. In particular, the SSD performance deteriorates due to garbage collection.