The short and easy System Usability Scale (SUS) has been found by Tullis and Stetson (2004) to psychometrically outperform other subjective scales including the renowned QUIS. Most SUS items seem related to learnability or memorability, along with a couple for efficiency. However, I wouldn’t try to break it into subscales; all items are highly intercorrelated suggesting this scale measures a single underlying construct.
I would doubt you can get a scale to measure each of Nielsen’s dimensions separately. A user can tell you if a product is “hard” to use, but it’s much more difficult for them to break it down further. They know it took a lot of work to do something, but was it because they couldn’t figure out an easier way (learnability)? Or maybe they had learned a better way on a previous task, but forgot it (memorability)? Or is that just the way it has to be (efficiency)? Users are not going to have sufficient information to make the distinction.
If you are specifically interested in each of Nielsen’s dimensions separately, then assess them separately and directly. You can measure learnability crudely through recording the number of errors or time between clicks, and precisely by how many trials it takes for users to learn the normative interaction sequence. For efficiency, after you train users to do the normative interaction sequence, record how long it takes them to do it. You can also get a pretty good answer analytically using something like GOMS-KLM. For memorability, bring the same users in a week or so later and compare their performance to that of the efficiency-measuring trial.
Like nearly all subjective scales, the SUS is primarily useful for comparing the overall subjective experience of different products. It’s hard to know what to make out of a single score without something to compare it to. These scales won’t tell what specific problems a product has or why it has them (e.g., to help you determine improvements). For that, qualitative observation and debriefing your test participants is best.