I used to have a really bad case of The Next Big Thing, but I've slowly started to come round to the idea of taking the time to study what already exists and consider where I might be the one who needs to learn and adapt. I've only just started with this train of thought, but I think this mindset is one critical of the process of doing things that are accessible and successful.
Someone once told me that to get anywhere you have to come up with a pie-in-the-sky idea that's absolutely crazy and then go for it. While taking that literally is a recipe for superfast burnout, it seems to me that that mindset tends toward system-based rather than goal-based motivation so might have some reasonable benefits for creativity and creative discipline. Not sure. Still figuring it out.
I definitely am interested in absorbing as much as I can. I've been figuring out how to build a tab/bookmarks/history-management extension for a while now, hopefully I get the courage to start (Chrome's APIs are so verbose and complex, and JavaScript requires so much boilerplate, I can't say I like it). But I currently have 652 tabs open that I need to bookmark and close, and something like 20k bookmarks that I need to tidy up (!), so it's on the todo list. Heh.
The first time I heard about hardware glitching was "Many Tamagotchis Were Harmed in the Making of this Presentation", https://youtu.be/WOJfUcCOhJ0. (I also just discovered and watched the update, http://youtu.be/mCt5U5ssbGU.) That was fun to learn about, but now I realize this sort of thing is widely applicable it's even more interesting. Thanks for the headsup! The concept of hardware glitching is something I've been interested in for a while actually.
It never occurred to me that I could implement a secure enclave myself. I thought you needed a secure processor for the design to even be viable. I'm only aware of things like the ORWL secure PC (https://www.crowdsupply.com/design-shift/orwl, https://www.orwl.org/wiki/index.php?title=Main_Page), which does use a secure processor (http://www.st.com/en/secure-mcus/st33g1m2.html) to manage the system's various security features.
It's mostly my complete domain ignorance, but I can't envisage a way to build a truly secure processor setup, mostly because of limited access to secure parts. A fast OTP microcontroller with enough space for a burnt-in key and the ability to interface with external Flash could work, but if I just used this for storage I/O going to another CPU, you could simply tap the bus lines to achieve untraceable information leakage.
The secure processor would need to deal with everything between keyboard input and LCD update, and only output encrypted data to the 2G radio. The chip I linked is only 25MHz, which would make for quite a limited device. It most definitely would work - I have an Ericsson MC218 that's that fast, and the EPOC OS (forerunner of Symbian) on it is incredible - but it would be much more accessible if the CPU were 250MHz or so instead. I'm not aware of secure processors that are that fast - and does the chip I linked even require an NDA to use? It doesn't look like it but I wouldn't be surprised if it did.
Ideally, I'd love for a way to use one of SiFive's RISC-V chips as the secure processor when they release their high-speed designs. But implementing secure boot on one of those would depend on both how the chip is designed (eg, boot sequence sensitivity considerations) and how the chip is physically constructed (I expect RISC-V chips with active meshes etc will eventually exist).
My pie-in-the-sky step-up from this basic concept would be to make a dual-chip system, with a secure CPU sitting alongside an off-the-shelf ARM CPU running Android. The secure CPU can take control of the screen and keyboard in such a way that the ARM CPU cannot attack (the keyboard would be easy - just route all keyboard I/O through the secure processor - but I fear that wielding the video side of things would be incredibly nontrivial to implement securely). Then when you wanted to do secure tasks you can simply tell the system to switch to the secure processor, which takes over the screen and keyboard until you tell it to return control to Android.
My ultimate goal would be a secure processor fast enough to capture medium-resolution video (something like 640x360 max, to begin with) from a camera module, and then play it back on an LCD, all without any sensitive data leakage (or depending on external processors that would require that). Ideally I'd like to go higher, but I think these are reasonable (beginning) expectations for a device that I would rather not put a GPU in. (Yes, crazy, but GPUs require NDA'd firmware, so the best ever case scenario I could manage is getting access to the BSP source and looking it over, but I'm most definitely not a security researcher, so I don't consider it viable. I can get away with Wi-Fi+cellular because the data going over that would already be fully encrypted with keys those chipsets cannot access, regardless of how malicious they are.)
Regarding the handheld not-quite-sure-what-it'll-be-yet thingy, the keyboard has been my biggest perplexion for a while. :) Tactile switches with low force actuation is one simple solution, but will never feel as professional as a proper rubber-dome actuator setup or similar. I've never used one, but the original Blackberry (pager) looks really close to what I want (in fact I've heard that thing runs on an 80386-compatible CPU - not sure of the manufacturer - and that there was once a devkit for it floating around and generally available). I wonder whether it uses a rubber actuator system or tactile buttons.
I completely understand your closing key point about actually manufacturing stuff though. I've gone for a very long time with just pondering and wondering, and no actual iteration, and I can't help but acknowledge that there is a lot of truth in the idea of "quantity over quality" - or more accurately our brain's ideas of "quality."
The idea that studying a subject with the notion that improving our understanding of that subject will make us better at it does hold true for a lot of areas and domains, but I think it tends to break down in a lot of the creative process. The process of making - whether that thing is something intangible like a piece of software, or a physical product - is generally something that must always be learned as a discrete subject nowadays. Unfortunately, this seems to be a rather hard idea to grasp, and there's a bit of a learning curve to it.
We depend on so many tools now, and those tools have developmental and process histories of their own that we need to appreciate in order to take the best advantage of those processes.
But our brains are likewise tools, and to use them most effectively we have to figure out how they work best. That process is a bit like jumping off a philosophical/psychological cliff :)
As for practically running off with any of these ideas and actually getting started with them, that's a ways off yet. I unfortunately don't have the budget for those things right now due to fun, expensive medical issues that make it impossible for me to get a job (yeah).