Replies

Definitely encrypt.

But I’m kind of at a loss how new “data” can coexist with overwritten data. On magnetic disks, you analyze the disk surface around and beneath the last written data for residual magnetism since every write does not follow precisely the same path or depth of previous writes and old magnetism can spread a little. How does something analogous happen to discrete capacitors?

You’ve got me. I was as surprised by this article as anyone.

But I’m kind of at a loss how new “data” can coexist with overwritten data. On magnetic disks, you analyze the disk surface around and beneath the last written data for residual magnetism since every write does not follow precisely the same path or depth of previous writes and old magnetism can spread a little. How does something analogous happen to discrete capacitors?

I had the same question, so I checked the paper that is referenced in the article (it's at Reliably Erasing Data from Flash-Based Solid State Drives).

It turns out that the concerns are not at the physical level; it's not the old problem of stray magnetic signatures that might be reconstructed. Instead, the problem lies in the fact that when you tell the SSD to write at a given block, and then later, you tell it to overwrite the same block, it doesn't actually do that. The logical block address mappings on an SSD are more sophisticated than most traditional hard disks.

The explanation in the paper is quite lucid:

The internals of an SSD differ in almost every respect from a hard drive, so assuming that the erasure techniques that work for hard drives will also work for SSDs is dangerous.
SSDs use flash memory to store data. Flash memory is divided into pages and blocks. Program operations apply to pages and can only change 1s to 0s. Erase operations apply to blocks and set all the bits in a block to 1. As a result, in-place update is not possible. There are typically 64-256 pages in a block (see Table 5).
A flash translation layer (FTL) [15] manages the mapping between logical block addresses (LBAs) that are visible via the ATA or SCSI interface and physical pages of flash memory. Because of the mismatch in granularity between erase operations and program operations in flash, in-place update of the sector at an LBA is not possible.
Instead, to modify a sector, the FTL will write the new contents for the sector to another location and update the map so that the new data appears at the target LBA. As a result, the old version of the data remains in digital form in the flash memory. We refer to these “left over” data as digital remnants.

The researchers did show that the built-in "Secure Erase" feature on some SSDs did correctly erase the entire drive; however, it was not present or failed on 8 of the 12 drives tested. Overwriting actually did work pretty well in many cases according to the paper, but it wasn't perfect.

Missed your post first go round. I have the same question as you. How can you have residual “ghost” memory in an overwritten binary system?