• Tin4 a 3x4 font (which allows hundreds of unique and legible Latin alike chrs).
• You are not trying to reinvent the wheel, like meek's PhoneStruct. Your Tiny3 is better IMO (more legible and standard).
• More glyphs, yes, but still no lowercase ! An update, perhaps ? (or you surrender in advance ? ;-)
• Femto 3x3 by Frodo7 : dead link. Any 2nd chance on FunkStrot ?
• Possibilities decrease a lot in 3x3, you're right.
• As I did not see Frodo7's work, I just can assert (from what I see in your), that : A can differ a lot, D a bit, E a lot, H not easily, I a bit, K a bit, O a lot, P a bit, R a bit, S not easily, T either, U a bit, V a lot, X not easily, Z either, 0 a lot.
• Yes, the smallest size for a Latin font, without minuscules or/and subpixels involved.
• I liked your effort to avoid the duplicates (though there are still a few).
• Now could you add the lowercase letters ?

B.t.w. when you said :
“This is my smallest font ever, 3x3 pixels”

Unfortunately, again and again (bad joke, Google !), you're absolutely wrong :
2x4 fonts are feasible too,
and they are even smaller(*),
and less legible than 3x3 ones.

* : Let's calculate to compare :
2x4 = 8 bits ; 3x3 = 9 bits.

By '2x4', I mean full ASCII, i.e. no duplicate, and the less illegible possible for this size, of course.
I'll release mine a.s.a.p. (it's finished), but please wait.

dpla

Google overlooks the 'my', sorry.

But please, can you provide us a 2x4 font now ?
I think you're trained enough :-)

Way too small - CMunk

@Håvar Henriksen (farside): sorry for the delay (which did not help you improve your 3x3 trials, unfortunately). I'm afraid you're still very optimistic (and not trained/daring enough for a 2x4 even in 2017?).

Please give us the serious competing LINK, my friend! :-))) (I cannot unveil my designs* before I have quite finished the typeface itself, in order to prevent more flawed fonts than we need; but if you can read my comments and tips on FS, you'll find the/my related solutions, I'm sure - if not, then see you again in a couple of decades…)

If it's not (at least) US-ASCII, it's not a font, I often repeat. Plenty of 2x4 glyphs exist, we know, on FontStruct, BitFontMaker, even on blogs nowadays, but never as an actual font, believe me from what I know (every time I read the contrary, I waste my time…). E.g. a single 2x4 letter is not a font; a set of duplicates is not a font; a very illogic design is not readable then not a font… I archive the most serious low-res pixel fonts, and your 'someone' looks like a biting HOAX to me :-)))

[P.S. since you posted your hint without a link on the same day as Umbreon126's comment, I really hope that you were not referencing only CMunk's Way too small, which is not a valid 2x4 design, not a valid US-ASCII either, but a very good trial all the same! The web does not limits to FS and its -young- creators… All right, as you can see, I 'timestamped' the precise date when I finished my 2x4, on this link.]

FS is a notorious toy/game (where 2x4 alike fonts might be understood as longer-to-design works by the neophyte or transient user?). I never said my 2x4 was quite readable, even less wholly legible, but if one needs a valid solution in strict 2x4, then I can provide the -to me- best 94 US-ASCII glyphs one can logically design (from many tests of mine, based on pixel art and coding, and more good practices in general). Moreover, as a font, it belongs to a family (of larger matrices). THIS is quite unique! (FS does not help at all, even with its cloning/renaming feature, you know…) Hundreds of hours of working time (as thousands of fonts)… perhaps it is the reason why you still have a lot not-so-necessary rules in your 2x4?

* How much would you pay to see/use my 2x4 US-ASCII?

OK, a few tips for you (just you!^^) if you still want to release it (the world smallest dot-matrix/bitmap/pixel/raster US-ASCII font on earth/galaxy/universe etc. ;-))

• I use a few hacks, but they are quite understandable after a short learning curve.

• I use at least one subset of a larger font (matrix), but unfortunately most -novice- fontstructors avoid it or just cannot see this set (which uses another hack, very logical too, and I'd say enough legible); without these glyphs, then, well… you cannot design a comple US-ASCII in 2x4, that's why I coud not release the 2x4 before 2014 (after my 3x3 fonts were sound enough).

• In your examples, you omit the cropping, amongst more fallback solutions and hacks (you might need to train even more, but I'd recommend you try some pixel art tricks, we never know if you still think the glyphs in their 'unnecessary entirety' - the main elements counting more than the secondary ones).

• My UC "A" is quite legible next to my LC "a", I needn't go further in detail.

• The same with "B/b" :-)

• “As you could see”: I saw nothing that invalidated my assertions, just 2 bad 2x4 glyphs of yours that only prove you limited your choices. When I comment (seriously) on this kind of binary-related topic, I provide ALL the possible glyphs, before discarding the less readable/legible ones. Again, my friend, if you had built the binary chart in 2x4, you'd have already seen the 2x4 glyphs I retained eventually (even though a very-low res design without any hack is not viable because of the missing segments etc.).

• “it's almost imposbile to make […] legible”: agreed. But coding-wise, there's enough room for the related 7-bit glyphs, so if you prefer 3x3, I have (the best ones - I compared in US-ASCII only), too.

• “There needs to be a hole in the middle on some of the characters”: oh, no! you too have the syndrom of the WYSIWYG… please see beyond! I remember such a comment on the The Bee's knees experiment (2009-09 Miha), where people cannot even imagine a flat LC "e"… do some pixel art, again, if you cannot understand our Latin glyphs without these elements of identification. Bottom line: incorrect assumption (in this specific field).

• I never cheat with the serious specifications in my typeface: matrices and readability are top-notch (i.e. for programming purpose, basically but not only). The legibility is what is supposed to be at this scale: very/extremely poor, although it depends on the glyphs and of some variable learning (the 94 glyphs have variable difficulties, but I am/was already proud of being able to provide a very coherent solution for most of them).

Do you need more -non-spoiling- proofs of feasibility in 3x3 or 2x4?

If not, you can just wait more, since I'll release the typeface or its script when I can… (A decent US-ASCII 3x3 being not even available on the net, you might understand the added value of such a modest, yet very time-consuming project.)

That's all, folks!