I'm reviewing our company's security procedures regarding cryptographic keys and master passwords, and also my private system at home. The key generation process is pretty much the same as that outlined in Recommended operational security for generating one's primary key pair?. Current procedures for paper backup and media for vault storage are more than sufficient for the operational lifetime of the material, which is at most three years for the certificates used for data exchange in the German health care system.
What I'm having trouble with is long-term storage, though, both for the paper-based part and the 'digital media for the bank vault' part. For several reasons in can be necessary to use certain keys a long time after their operational life has ended, even decades later. For example when digging in old archives, or when trying to read a really old PGP message.
And it is the 'decades later' part wherein lies the rub. I couldn't find good data for the durability of printed matter, beyond various recommendations/requirements to use dot-impact printers or black-and-white laser printers. Laser printers offer higher quality and hence higher possible data densities but from the info I found I couldn't say whether their printouts really last longer than those of dot-impact printers or not...
As regards digital media, those in current use all have an expected durability of only five years or less, except for flash devices and hard disks to which a somewhat longer - though unspecified - durability is attributed. Our only long-term durable media - MO disks - had to be scrapped and shredded when we retired the last available MO reading device.
Based on my unscientific Googling I'm currently favouring USB sticks for digital storage, as they seem to combine long-term data retention (internally) with a good prospect of long-term accessibility via the USB plug/bus interface (externally), but I'm not quite satisfied with the scarcity/quality of corroborating information that I've found so far.
Another consideration is that printed matter can be inspected visually but digital media can not. There's no indicator that the data might be hanging on only by the tips of its fingers (figuratively speaking) and the next stray cosmic ray might blast the last 'good' electron from its perch and thus cause data loss in a cell. Digital media might be close to a catastrophic collapse without anyone being able to detect that fact.
I couldn't find recommendations as to a possible refresh write every n years or so, or regarding suitable storage formats with high redundancy (ECC/Reed Solomon). Given the size discrepancy between key material and storage capacities, even a hundred-fold size increase would be perfectly acceptable. It might be possible to put the redundancy features in tools like RAR to good use here. However, I have no experience in such matters, and procedures/schemes designed by amateurs pondering first principles tend to be cr*p in actual practice...
I'd appreciate pointers to studies, papers and articles regarding the secure long-term (vault) storage of key material, both on paper and on digital media. I'd also welcome pointers regarding practice and experience in this context. The focus would be mainstream tools and technology of the sort accessible to a small company or a private citizen like me, not specialised tech with price tags of four digits or higher.
Note: I'm also interested in the practicalities of paper backup, in particular tools that allow adding error-detection and error-correction capabilities similar to what RAR does for archives (i.e. dial-your-desired-security), and perhaps encodings that are more efficient and/or robust in the face of OCR than, say, base64.
Update: durability of print is the subject of ISO 11798 which helps with googling related information. See Erstellung von alterungsbeständigen Personenstandsunterlagen at the Landesarchiv Baden-Württemberg (in German), for example, or HP Supplies Print Permanence and Durability. It seems that laser printers have no problems passing the tests (whereas ink jets have to work at it, hard).