Samsung-Backed Researchers Develop a Ternary Semiconductor (zdnet.com) 164
"Future semiconductors may perform logic with 0, 1, or 2 instead of the current binary system of 0 and 1," reports ZDNet:
A South Korean research team has successfully realised an energy-efficient ternary metal-oxide semiconductor on a large-sized wafer. Professor Kyung Rok Kim of UNIST's Electrical & Computer Engineering Department and his team successfully created a semiconductor that operates in a ternary logic system instead of the current binary...
Using the ternary system of 0, 1, 2 lessens the amount of information semiconductors need to process and does it faster, resulting in less power consumption, the team said. It will also help in miniaturising chips further. For example, to express the number 128 in the current binary system, 8 "bits" will be required. With the ternary system, only 5 "trits" will be required....
Samsung Electronics has been backing Kim's research since September 2017 via its Samsung's Science & Technology Foundation, which offers grants for promising technology projects. Samsung is currently verifying the technology at its foundry business-run fab.
Using the ternary system of 0, 1, 2 lessens the amount of information semiconductors need to process and does it faster, resulting in less power consumption, the team said. It will also help in miniaturising chips further. For example, to express the number 128 in the current binary system, 8 "bits" will be required. With the ternary system, only 5 "trits" will be required....
Samsung Electronics has been backing Kim's research since September 2017 via its Samsung's Science & Technology Foundation, which offers grants for promising technology projects. Samsung is currently verifying the technology at its foundry business-run fab.
So logic becomes... (Score:5, Funny)
Yes, No and Maybe... Or True, False, and IDK...
Quaternary? (Score:2)
Yes, no, maybe and maybe not.
Re: (Score:2)
Re: So logic becomes... (Score:1)
Re: (Score:2)
that's for the future, it's been prophesied by Star Trek ;-)
Re:So logic becomes... (Score:5, Funny)
True, False, and FILE_NOT_FOUND.
Re:So logic becomes... (Score:4, Funny)
Abort, retry, fail.
Yes, No, Fake News (Score:2)
Re: (Score:2)
I thought that abbreviated to tit.
Re: (Score:2)
Abort, retry, fail.
Found printed on a pregnancy test?
Re:So logic becomes... (Score:4, Informative)
Logic is still binary, that's just math. With this, storage is multi-level and undergoes a conversion process to pack and unpack.
Re: (Score:2)
Re: So logic becomes... (Score:1)
Re: (Score:2)
As useful as binary logic is for a subset of problems (generally the ones with absolute defi
Re: (Score:2)
No, real-life logic is ternary. True, false, and unknown/cannot be determined. Goedel's incompleteness theorems [wikipedia.org] proved that no matter what logic system you come up with, there will always be cases which are unprovable in that logic system. e.g. If you limit your logic to binary true/false, is the sentence "This sentence is false" true or false? Binary logic cannot deal with that simple case - it is incomplete.
That is not an example of Godel's Incompleteness Theorem. The fact of the matter is you can't give an example. There are true statements that cannot be proven to be true, but you would need to know it's true to give an example.
The example you gave is inconsistent with binary logic, I suspect because the self reference in that statement makes it like a set which contains itself (which doesn't exist). A true but unprovable statement is a consistent logical statement, otherwise it could not be said to be true
Re: (Score:2)
Sorry, I mean contemporary logic is still boolean. Ternary logic is not going to be used in general purpose computing.
You may find small accelerator sections on microcontrollers with ternary logic already, and it is thought it might make some of the so-called "deep learning" processors a bit more compact if ternary logic were supported in their convolution matrix operations.
It's not clear if Samsung has much interest in the AI or even the compute industry. I guess we'll wait to see if Samsung makes a super
Re:So logic becomes... (Score:5, Funny)
True, False, Uninitialized
There are 10 kinds of people: Those that think that this is a binary joke, those that understand ternary and those that don't.
Re: (Score:2)
C# actually has that: the nullable bool. It can be true, false or null.
Re: (Score:2)
That;s the compiler's job, not the language's.
Re: (Score:1)
It's actually 0 1 and another 1 in that sense. 2 becomes the other yes. It's b
Re: (Score:2)
True, undetermined, and False. Tenery Logic become complicated quickly, with the binary options, plus quite a few weird ones. It can make math easier if you use assign values of 1, 0, and -1, but not that much easier. And while you save on the data path, functional units and logical operators don't actually shrink much.
I also suspect conversion from binary representation is hell, with no way being much better than cascaded modulo operations.
Re: (Score:2)
"Yes, No and Maybe... Or True, False, and IDK... "
2 is an alternative fact of 1.
Re: (Score:2)
True, False, and Your Credit Card is Not on File.
Re: (Score:2)
Expect the Bro-Programmer culture to be quickly replaced by all female teams. This is literally how my wife's mind works.
Re: (Score:1)
Imagine an NVMe card that has the trinary decoders built into the chip. It would be transparent to the rest of your computer. I don't think people are suggesting we walk away from Boolean algebra and 8-bit bytes.
Re: (Score:2)
This is just like that useless TV that had the colour yellow in it. No one will use it, because no videos support it, no decoders support it, it isn't filmed for it, and so forth.
That TV wasn't useless. Most panels really do have a gap in color gamut. Even though the color can be represented as R, G, and B values doesn't mean that the panel can't show the colors more accurately by adding yellow in place of R+G.
another angle on binary (Score:2)
At first I was thinking "why don't they just convert to binary at some point to keep it standard - increase storage or data density but not improve processing."
But then I realized a power of 3 will never fit exactly into powersof 2, so the extra information ("possibilities") would be wasted.... or maybe not? If you just flat out replaced 8 bits with 8 ... terts? (is there a name for a trinary bit?) then you'd have at least two surplus terts that could be used for checksum or error recovery, with maybe no
Re: another angle on binary (Score:2)
The word you're looking for is "trit".
Re: not again (Score:2)
How do you throw out "the right" trit when there's only "one" trit that supports 0v, 5v or 10v to mean 0, 1 or 2 respectively?
You have one apple in your sack. Throw out "the right" apple. How do you do that?
re-educate (Score:1)
Re: (Score:3)
Re: (Score:2)
There is already a lot of flash memory on the market with up to 16 levels per cell.
Re: (Score:2)
And now they can do some of the control logic in multi-level natively, maybe the error correction. It's a niche application.
Re: (Score:3)
How do gates work with ternary? This bit XOR? This doesn't work with discreet mathematics at it's crudest level.
The way XOR works in binary is simply a definition, nothing more. Same for every logical operation. So the way XOR would work in trinary is the exact same way: however we choose to define it. At it's crudest level, binary logic operations map pretty well to operations on sets, so you might want to ensure your new definitions are backwards compatible. Or not.
Re: (Score:2)
How do gates work with ternary? This bit XOR? This doesn't work with discreet mathematics at it's crudest level.
It works fine, as long as you don't do it in public and don't tell anyone.
not much detail (Score:2)
Article doesn't say much, and the real paper is behind a paywall.
The real challenge would be to develop the actual logic, rather than just the semiconductor structure. How are you going to build a hardware multiplier that works in ternary, for example ?
Re: (Score:2, Interesting)
Same author (Kyung Rok Kim) has an 2017 IEEE publication titled "A Novel Ternary Multiplier Based on Ternary CMOS Compact Model", for instance.
Re: (Score:3)
http://soc.unist.ac.kr/~www/Pa... [unist.ac.kr]
Thanks for the link! (Score:2)
This is pretty cool; it allows scaling the number of gates at least ~60%, so a lot more can go on a die.
I think noise immunity will be the big problem; what happens when a cosmic ray hits this structure?
If it can be predicted the same as current processes, it will be a hit.
Balanced Ternary (Score:4, Interesting)
Hopefully it will use Balanced Ternary encoding https://en.wikipedia.org/wiki/... [wikipedia.org].
There was some experiment in ternary logic in the beginning of computer history, but it made circuit more complex, so it stopped soon.
But I suppose it will never be common, because it's too different from current architectures.
Re: (Score:2)
But I suppose it will never be common, because it's too different from current architectures.
This is also a very valid point. We carry many compatibility leftovers since the 1950s (i.e. certain terms that would no longer make sense if introduced today, certain alignments such as block sizes of 512, etc.) so... Interfacing with preexisting practice is a great deal.
Of course, if we are talking about devices to hold information (i.e. flash drives), and the controller can make the translation... Does not matter much!
Re: (Score:2)
From what I can deduce from the paper's abstract, and some mention in Wikipedia, it's unbalanced ternary.
Balanced ternary is not particularly difficult to achieve, I remember running some SPICE simulations on ternary CMOS many years ago. But I'm not an EE, so I'm not sure if they were actually physically feasible.
Still, unbalanced ternary could be very useful if we can run the same computations with less transistors. And for some really specialized tasks, like GPUs or neural networks, backwards compatibilit
Re: (Score:2)
From what I can deduce from the paper's abstract, and some mention in Wikipedia, it's unbalanced ternary.
If the process can handle 3 different states, surely it can be used to represent any choice of three, including {-1, 0, 1}, {a, b, c}, or {male, female, attack helicopter}.
Re: (Score:3)
There was some experiment in ternary logic in the beginning of computer history, but it made circuit more complex, so it stopped soon.
Per the Wikipedia page to which you linked, balanced ternary "cuts down the carry rate in multi-digit multiplication, and the roundingâ"truncation equivalence cuts down the carry rate in rounding on fractions. The one-digit multiplication table has no carries in balanced ternary, and the addition table has only two symmetric carries instead of three." And because integ
Re: (Score:2)
Yes it's simpler at logic port level, but ternary logic at the time require more transistor per logic port, so in the end was more complex.
I can't find the original citation, but that is what I remember I had read.
Re: (Score:2)
When you look at how languages like C# are actually used it's not hard to see how ternary might be used to execute them. The language goes a long way to hide implementation details like the binary types used.
They even have a ternary type built in: the nullable bool. It can be true, false or null.
Re: (Score:3)
But I suppose it will never be common, because it's too different from current architectures.
It will never become common because there is always a price to pay for putting more information on a wire, in terms of lowered temperature, noise or voltage tolerance.
They got the name wrong (Score:5, Funny)
Binary digIT = BIT, so Ternary digIT = um, . . .
Re: (Score:3, Funny)
Yes, No, TITS UP.
Re:They got the name wrong (Score:4, Funny)
Or...
2 levels: Bi-nary digiT = Bit :-P
3 levels: Ter-nary digiT = Tert
4 levels: Quater-nary digiT = Quatert
5 levels: Qui-nary digiT= Quit
Wake me up when we've got quaternary digits. Being 2**n, they would be easier to interface with binary and therefore I think the most useful.
Re: (Score:2)
Or...
2 levels: Bi-nary digiT = Bit 3 levels: Ter-nary digiT = Tert 4 levels: Quater-nary digiT = Quatert 5 levels: Qui-nary digiT= Quit :-P
Wake me up when we've got quaternary digits. Being 2**n, they would be easier to interface with binary and therefore I think the most useful.
I'd make the 4-state digit a Quart. And then you can have a Pentary digit and make it a Pint, just to screw people up.
Re: (Score:2)
Most folks avoid that problem by calling it a Trinary Digit--a "trit"
Re: (Score:2)
Re: (Score:2)
MMm Mmm MMm, you make me wish I had three ANDs...
Reliability? (Score:3)
This is not such a new development... And I'd like to see a clearer reasoning behind it (yes, I know, RTFA). Issue is, we have developed binary systems because the real world is noisy. It's easy to clearly separate a "high" and a "low" value. Label them 1 and 0, if you wish. So, it was simpler to train tech-savvy humans to understand electronics-friendly discrete values.
If a ternary electronic component is introduced, then a clear and reliable way to clearly distinguish between low, middle and high will be needed. We still have error detection and correction hardware, because electronics... Well, they are analog and funky.
Again, I won't say "this won't fly", but rather, "we should read into the motivation for doing this". I really really really doubt it's in order to more densely pack memory.
Re: (Score:2)
Re:Reliability? (Score:4, Interesting)
TFA is just a news note. The "real" article is paywalled, so I got only to the abstract - and both convey very little additional information to what is already reported at /.
It strikes me as intersting that it mentions ternary _logic_. Do they mean making a ternary ALU? Ye,s the silicon footprint for that would be smaller, but the complexity would surely rise (maybe even obliterating the savings!)
Re: (Score:2)
There's a link above in the comments; it is more efficient.
Read the paper. :)
Re: (Score:1)
Re: (Score:2)
Long ago, one of my professors in my digital logic class said that mathematical analysis could prove that the most theoretically efficient
Re: (Score:2)
TriINTERCAL anyone? (Score:1)
Finally a machine suited for a native TriINTERCAL implementation.
Contrived example, can't see it going anywhere (Score:2)
My prediction is except a few niche cases this is going nowhere since it is a lot of change in the way things are done an
Re: (Score:3)
Important benchmarks will be double precision floats, which could be reduced from 64 bit format to 40 trits, or low precision math (e.g. for machine learning) which could be optimized for new format.
Re: (Score:2)
The value 128 may take 8 binary bits whereas compared with 5 ternary bits, a 3 bit advantage but that is rather contrived. For example 127 needs only 7 binary vs 5 ternary bits, only a 2 bit advantage and 8 binary bits gets you to 255 which takes 6 ternary bit, still only a two bit advantage. Using contrived examples is not going to win them a positive perception from people like me.
Worse, a ternary bit is only an improvement if it's in the same space and power envelope at scale. AFAICT, the article deals with some aspects of this, but it doesn't really say whether it's generally adaptable to the materials in use today. As a for-instance of where this might go wrong, what if the third voltage state doesn't scale down the same way as the primary two voltage states?
Re: (Score:2)
Russians did it first (Score:5, Informative)
Way looong ago!
https://en.m.wikipedia.org/wik... [wikipedia.org]
Re: (Score:2)
Why stop at 3? Nature has been using 4-level storage and processing (via the 4 DNA/RNA bases) forever.
Re: (Score:2)
Because e is the optimal value.
Re: (Score:2)
okay why is e the optimal value? I've heard that said so many times
Re: (Score:2)
The optimal base x for a number N is one that maximizes (N/x)^x.
Re: (Score:2)
why? I'm fine with math but I don't see what that optimizes. And then the follow up question is why does optimizing X for property Y help a computer?
THe english did it in 1940 (Score:2)
https://ieeexplore.ieee.org/do... [ieee.org]
Re: THe english did it in 1940 (Score:2)
1840
{-1,0,1} is the natural basis (Score:5, Interesting)
One of the advantage of the basis {-1,0,1} instead of {0,1,2} is that no sign bit is demanded. Operations like rounding and truncating are the same.
Re: (Score:2)
One of the advantage of the basis {-1,0,1} instead of {0,1,2} is that no sign bit is demanded. Operations like rounding and truncating are the same.
OTOH, number representations become a bit weird, because positions can have negative values, so numbers are constructed by subtracting as well as adding. For example 5 in balanced ternary is represented as +--, where + and - denote positive and negative 1, respectively. +-- = (1 * 9) + (-1 * 3) + (-1 * 1) = 9 - 3 - 1 = 5. In the unbalanced basis, {0, 1, 2}, it's a more typical base conversion that yields 12 (base 3) = 1 * 3 + 2 * 1 = 3 + 2 = 5.
This probably isn't actually a problem, but it's definitely
closer to perfect efficiency but (Score:1)
Not new (Score:1)
This is not new. I knew IBM did it in the 90s I believe but the history is much older.
https://en.wikipedia.org/wiki/Ternary_computer
Howabout 4, like DNA? (Score:1)
Re: Howabout 4, like DNA? (Score:2)
Why binary? Wouldn't some other base be easier.
Well, yes, but....
https://youtu.be/thrx3SBEpL8 [youtu.be]
Re: (Score:2)
Great idea, but needs extending (Score:2)
The real goal should be bits with infinite states -- talk about space savings! We could call it "a N/A log" or something catchy like that...
Russian computers anyone? (Score:2)
https://wikivisually.com/wiki/... [wikivisually.com] ... don't recall if any of the Elbrus systems used trits as well.
Re: (Score:2)
https://www.jstor.org/stable/2... [jstor.org] for some background on why folks might want to do this.
Trits? (Score:2)
Since "bits" was derived from "binary digits", shouldn't the corresponding term for "ternary digits" be "tits"?
Re: (Score:2)
Well (Score:2)
That's odd.
Nightmare for harddisk sizes (Score:2)
will that break my programs where I used.. (Score:1)
n<<1 instead of 2*n
magnetic memory (Score:2)
back in the 70's we used magnets - north, south, none for ternary logic memory.
like core memory on steroids.
Ternary (Score:2)
Ternary Arithmetic (Score:2)
http://homepage.divms.uiowa.ed... [uiowa.edu]