Choosing the Insertion Point in the Datapath
A self-encrypting drive can apply its transform in one of three places: host-side software, the controller's internal datapath, or closer to the media. Enterprise designs take the second option, placing the AES-256 engine between host DMA and media encode so that data is transformed inside the hardware pipeline in both directions. Bandwidth is the first reason: the portfolio's planned peak sequential read is 14,500 MB/s, and any crypto path requiring CPU involvement becomes the bottleneck at that rate. The second is latency budgeting — an in-pipeline hardware engine works at line rate and contributes a fixed number, not a load-dependent source of jitter.
Keys Never Leave the Controller
The media encryption key is generated and used inside the controller and is never exposed across the host interface. A key derived from user credentials only wraps that media key; the wrapped blob lives in the controller's secure region, survives power loss, and must be unlocked through authentication after power-on. The hierarchy has one immediate benefit: changing a passphrase requires only a re-wrap, not a rewrite of every byte on the drive.
OPAL Locking Ranges Are Not Namespaces
TCG OPAL locking ranges are carved out by LBA span, each with its own credentials and lock state. NVMe multi-namespace support partitions at the logical device level. The two can be layered, but their boundaries must be aligned during configuration; otherwise a namespace straddles two locking ranges and comes up half readable after power-on. Combined with Secure Boot, the firmware image itself sits inside the signature verification chain, so encryption cannot be bypassed by substituting firmware.
Mutual Constraints with Compression and Erase
Ordering is not negotiable. Transparent-compression models must compress before they encrypt — ciphertext entropy is near random and compression ratio collapses to nothing. Secure erase draws on the same key hierarchy from the other direction: once the media key ceases to be valid, the ciphertext on the drive is no longer decryptable, which is a direct dividend of keeping keys inside the controller. All three capabilities share one key management design, so a change in any of them forces re-validation of the other two. These specifications and capabilities are planning targets today; as part of the enterprise feature baseline they are being brought up item by item on the engineering platform alongside the controller and firmware plan.