I'm really just a student learning this stuff for my own enjoyment.

While reading through one of the texts, it mentioned how logical gates can be controlled by diodes, but that diodes of different materials have different minimum voltage levels. It occured to me that this could be used for computing with more than the two values of 0 and 1.

It then also occured to me, that if positive and negative voltages could be used as well, you could nearly double the number of possible logic values.

As I'm nowhere near having a full education on the topic, I'm not sure how viable either of these are for practical use, nor for how a computer with more than 2 values would compare to a binary computer in terms of speed. I some some 3 value computers exist, but haven't found out how those represent the third value yet.

Anyway, I'm curious how viable any of these ideas are, or possibly where to find out the relevant info.

 

Welcome to AAC!

A concept related to this is used in Multi Level Cell FLASH memory to store multiple bits of data in a single memory cell. The original design stored information by depositing a "lot" of electrons on a floating gate. The presence of sufficient charge represented a '1', and less than that was a '0'.

In MLC, 4 charge levels were defined so 2 bits of information could be stored in a single cell. I think that has been extended to 3 bits.

Accomplishing 2 bits required precise control of the number of electrons as only about 10K separated each threshold.

Doing that for general digital circuitry isn't practical because the required error detection and correction clrcuitry would be costly.

 

http://www.bing.com/entityexplore?q...3d0a-9dae58a5bd21"&eeptype=Entity&FORM=SNAPST

 

Multi-level combinatorial logic has been investigated for many decades, but the increase in circuit complexity has always made it impractical. Storage is another matter, as above, and the flash memory people *love* multi-level stuff. So do the data communications crowd. Almost all of the Ethernet signalling protocols use more than two voltage levels. Modems do a similar thing by having a constant amplitude, constant frequency tone have multiple possible phase shifts.

ak

 

As AK noted, in typical logic/computer circuits the added circuit complexity to handle more than 2 levels of logic signal means you do not gain any advantage is processing power versus chip area (it could actually give less processing power versus area), and that's by far the overriding factor in determining whether multilevel logic would be used.

 

As AK noted, in typical logic/computer circuits the added circuit complexity to handle more than 2 levels of logic signal means you do not gain any advantage is processing power versus chip area (it could actually give less processing power versus area), and that's by far the overriding factor in determining whether multilevel logic would be used.

I can see that for some applications size would be an issue, but we've miniturized things so much that I don't really see why it would be a general/universal concern. We even get computers, phones and tablets in particular, so small that some people actually want bigger versions.
Is there some reason why the industry thinks things are still too large, or is it simply that size has never had it's priority looked at and adjusted?


Personally, I figured that if multilevel logic was ever used for computing, it would be for speed, computing ability (like three state machines easier handling of negatives), or memory space reasons (or a vastly different form of computing that naturally lends itself to more states). I'd certainly trade a bit of size for speed.

@dl324
Sp how does storing a set, and small, number of electrons relate to controlling volts? I would expect these to be vastly different, with volts easier to control, particularly since the margins between minimums can be increased, though at the expense of power. Can't mosfets output pretty precise voltage levels more than adequate for .4 volt margins?

As far as things being too costly, it is that focus on monetary gain and savings that is slowly killing the human race, thus I care not about the cost. (after all, it isn't business that drives the greatest human triumphs.)

 

Is there some reason why the industry thinks things are still too large, or is it simply that size has never had it's priority looked at and adjusted?

The complexity and cost of processing a single silicon wafer is almost impossible to imagine. a 1% increase in packing density is thousands of dollars per minute.

Personally, I figured that if multilevel logic was ever used for computing, it would be for speed, computing ability (like three state machines easier handling of negatives)

But that is a small percentage of the total workload of a general purpose processor. For any computing of all positive numbers, combinatorial logic, list processing, etc., the third state is not used (in your example) and thus is just a size, power, and cost and reliability headache.

(after all, it isn't business that drives the greatest human triumphs.)

In tech, business drives *all* triumphs. I'm not saying that is healthy or right, just true.

ak

 

I can see that for some applications size would be an issue, but we've miniturized things so much that I don't really see why it would be a general/universal concern. We even get computers, phones and tablets in particular, so small that some people actually want bigger versions.
Is there some reason why the industry thinks things are still too large, or is it simply that size has never had it's priority looked at and adjusted?

It's not the physical size of the end device that is so much at issue as it is the cost of the silicon. The mask set to make to make a wafer is up in the millions of dollars now. To make an economically viable product you must get as many chips from each wafer as possible. Part of that is physical size of the die, but a big part of it is how many of those die are good (the yield). Not only would multi-level designs increase the size of the die, but it would also significantly impact the yield. It would also greatly increase the cost of processing the wafers because your process has to be in much better control. Just look at the cost differences for running a design on a logic process versus an analog process that are nominally identical and you will get an appreciation for the impact.

Personally, I figured that if multilevel logic was ever used for computing, it would be for speed, computing ability (like three state machines easier handling of negatives), or memory space reasons (or a vastly different form of computing that naturally lends itself to more states). I'd certainly trade a bit of size for speed.

There's no guarantee that you would get any increase in speed at all -- it might likely slow things down. My understanding is that mutli-level Flash cells are considerably slower than single-level cells and they are also nowhere near as reliable. But they pack a lot more data into the same area and that is the driving factor for many applications.

As far as things being too costly, it is that focus on monetary gain and savings that is slowly killing the human race, thus I care not about the cost. (after all, it isn't business that drives the greatest human triumphs.)

This is an extremely naïve point of view -- and one which I guarantee you only give lip service to. Would you have the same computer you have now if it cost $15,000? No. Would you work for the same employer if they only paid you 1/10 what you currently make? No. Would you eat the same food if it cost ten times as much? No. See, you do care about the cost -- at least when it impacts you.

 

Would you have the same computer you have now if it cost $15,000?

For a typical mid-range desktop system ($500), that is only 30 times the current selling price. I think a better estimate of the true cost of a desktop system when not offset by the gigundous manufacturing scale cost savings is at least 50x. 100x would be $50,000, and that still might be low. For those who remember, think about the cost of a PDP-11/70 back in the 70's. In "computing horsepower", that system is maybe 10% of what is on my desk.

ak

 

For a typical mid-range desktop system ($500), that is only 30 times the current selling price. I think a better estimate of the true cost of a desktop system when not offset by the gigundous manufacturing scale cost savings is at least 50x. 100x would be $50,000, and that still might be low. For those who remember, think about the cost of a PDP-11/70 back in the 70's. In "computing horsepower", that system is maybe 10% of what is on my desk.

ak

Yeah, I started to point out how today's computers are significantly more powerful, perhaps orders of magnitude, than multi-million dollar mainframe computers fifty years ago.

Today's electronics are so powerful and so cheap largely (almost entirely) due to businesses being extremely sensitive to cost concerns. And businesses are so sensitive to cost because their customers, whether it be another multi-billion dollar corporation or some soccer mom, are so sensitive to cost.

 

I'm really just a student learning this stuff for my own enjoyment.

While reading through one of the texts, it mentioned how logical gates can be controlled by diodes, but that diodes of different materials have different minimum voltage levels. It occured to me that this could be used for computing with more than the two values of 0 and 1.

It then also occured to me, that if positive and negative voltages could be used as well, you could nearly double the number of possible logic values.

As I'm nowhere near having a full education on the topic, I'm not sure how viable either of these are for practical use, nor for how a computer with more than 2 values would compare to a binary computer in terms of speed. I some some 3 value computers exist, but haven't found out how those represent the third value yet.

Anyway, I'm curious how viable any of these ideas are, or possibly where to find out the relevant info.

I can see that being used for storage of data but counting, math or logic operations might be difficult.

 

I can see that being used for storage of data but counting, math or logic operations might be difficult.

Counting and math probably wouldn't be too difficult -- these concepts are distinct from the number base used. Many algorithms can be highly optimized to exploit base-2, but that is not to say that at least some of them couldn't be optimized to exploit some other number base even better.

The logical operations may be a different matter. Multi-valued logic systems have been around for a long time and if there ever comes a day when the benefits of multi-valued computation justify its adoption, we will probably figure out a workable way to use it pretty quickly and then see those abilities grow rapidly. But the issues inherent in the logic may prevent those benefits from ever reaching that level of viability in the first place. Just consider the shear number of operators that a three-level logic system has. With a two-valued logic system there are four possible unary operators and sixteen possible binary operators. In a three-valued system there are 27 possible unary operators and 19,683 possible binary operators. While this superficially indicates the expressive and computational potential of such logic, consider the design validation issues when it comes to ensuring that the system is going to behave as desired under all cases.

 

Sp how does storing a set, and small, number of electrons relate to controlling volts?

Q=CV.

Since the capacitance of a bit cell is fixed, the number of electrons deposited on a floating gate will determine how much the threshold voltage of the transistor will shift.

Can't mosfets output pretty precise voltage levels more than adequate for .4 volt margins?

In digital circuits, the transistors are either on or off. Voltage margin is used to account for differences in MOSFET threshold voltage and provide noise immunity.

The storage element in FLASH memories is analog. They use sense amps (comparators) and reference cells to differentiate between the voltages that correspond to programmed vs unprogrammed. In the case of MLC, there are 4 different voltages. The charge representing 01 and 10 need to be controlled more tightly than the charge representing 00 or 11; some residual charge is allowed for an unprogrammed bit.

 

===Caring about costs

It seems I was misunserstood.

A business seeking only to make money has concern for any and all costs and worry about making the minimum viable product plus a little bit extra to be above their competitors.

A group seeking to make a thing the best it can be on the other has very different concerns and indeed first focuses on making a rough draft design that is the best it can be then adjusts the design to stay close to that peak potential while allowing for more resource efficient techniques in crafting. Thus, cost issues come up later in the process and it is sought to minimize the effect these cost saving techniques have on the final product's abilities.

As you can see, a business seking profits vs a group seeking to reach full potential are opposite in design priorities. Busunesses don't just reduce costs, they reduce the product capabilities to the bare minimum that keeps customers, not just because it reduces costs, but also because they make greater profits by making improvements in steps. Do you think businesses would have released petabyte thumbdrives at first even if the costs were similer? To do so would be stupid from the viewpoint of making money, even if the costs were identical.

A group seeking to make the best of something would have released petabyte thumbdrives at the very beginning because seeking to make the best of something has priorities very different and quite contradictory to profit-seeking.


My handling of costs in terms of what I buy in the real world is pure strategy of me alone agaist those who actively want to take all my wealth regardless of the cost to me and my livelihood. That is not the way it needs to be though. There is no need to have an economy based on currency or the amassing of wealth, but these types of economies are the most advantageous for those who want to maximize the majority of wealth into the hands of the few. An economy built to maximize the enjoyment of life by the majority of a group's members would use a vastly different economy. And such a thing is possible. The Oneida community did very good and only the pressures from outside the group (at the wrong moment) is what broke them up.

I might be a bit idealistic, but if I'm going to spend time designing something, I'm not going to design a product to make a profit, I'm going to design a product to be the absolute best it can be, and if I get the chance to make it in numbers, then and only will I look at costs and I'll be minimizing the impact of the cost efficiency techniques on product performance. I want to make and see the potential of what can be achieved, not see how big I can make my bank account. I want to go to the moon, not conquer and take from those around me.
So no, I don't pay lip service to caring about costs. When it comes to designing something, I don't care, because I don't design things to make a profit, I design things to be the best that can be achieved, whether I'm designing it for fun or for real.

=== Computational ability

I can easily see why better handling of negative numbers is not always worth it, depending on the circumstances (clearly some cases are worth it as trinary computers exist), but negative numbers was just an example from the real world. There might be other advantages that can be found, particularly for more levels, perhaps in processor control schemes, error detection, security processes, scientific number analysis/crunching, etc.

===multilevel logic value operators
While multilevel logic has exponentially more operators as the number if levelw go up, this is not really an issue. No where near all the operators need to be used or even developed, you just have to develop the ones with the inputs and outputs required without enumerating all the possible operators. I.E. if you don't need an operator that outputs 0 regardless of inputs, then don't waste time on it.
===multilevel flash memory
First, I'd expect the need to translate incoming multi digit bits (10) to the single multilevel digit cells (2) would be a significant factor in speed. Of course, if you didn't need to translate, then how fast would it be?

Also, memory cells need to hold their state over the long term, while processing is continuous instead. Processing also would be using voltage to select from multiple paths rather than setting values in one spot. Perhaos I'm just ignorant in this case, as I have only started lookjng into things at this level, but having a .4 volt signal go one way or a .7 volt signal go another way seems far easier to me than translating binary into multistate memory cells, particulary if they need to hold their state without a power source.

 

===Caring about costs

It seems I was misunserstood.

A business seeking only to make money has concern for any and all costs and worry about making the minimum viable product plus a little bit extra to be above their competitors.

A group seeking to make a thing the best it can be on the other has very different concerns and indeed first focuses on making a rough draft design that is the best it can be then adjusts the design to stay close to that peak potential while allowing for more resource efficient techniques in crafting. Thus, cost issues come up later in the process and it is sought to minimize the effect these cost saving techniques have on the final product's abilities.

As you can see, a business seking profits vs a group seeking to reach full potential are opposite in design priorities.

What is your basis for claiming that these are how businesses seek to operate. You are setting up a straw man -- YOU are defining how YOU want to believe businesses operate for the purpose of then drawing the conclusion YOU want to believe.

My handling of costs in terms of what I buy in the real world is pure strategy of me alone agaist those who actively want to take all my wealth regardless of the cost to me and my livelihood.

In other words, when YOU care about costs, it is only because of all those evil companies out there. But when THEY care about cost, it is only because they are evil.

 

For a student you seem to be long on opinion and short on knowledge and understanding. Maybe your background is in a different field where esoteric thoughts are an accepted part of scholarly debate. In general extraordinary claims require extraordinary proof. I don't know that you have offered us anything that would be accepted as evidence.

 

Ternary logic has been done and was deemed successful but abandoned in favor of binary.

http://www.computer-museum.ru/english/setun.htm

If you don't know this bit of computer history, now is a good time to read about it.

 

For those who remember, think about the cost of a PDP-11/70 back in the 70's. In "computing horsepower", that system is maybe 10% of what is on my desk.

I bet it's less than 10%. I haven't looked it up but I bet it's at least 1000X increase since then. And I remember PDP11.

 

I bet it's less than 10%. I haven't looked it up but I bet it's at least 1000X increase since then. And I remember PDP11.

And the PDP-1, PDP-5, PDP-7, PDP-15, PDP-8, and PDP-10.
BTW the PDP-7 had a small role in the development of C
https://www.quora.com/What-language-was-the-first-C-compiler-written-in

 

Tri level logic did exist ,
and thats different to Tri state ,

yes there were three logic levels.

Some of the ex pander chips still use it so two pins can select 9 different states / address's .

problems were

a) No equivalent to the Boolean logic reduction / karnaugh map, so logic design was slow and risky,

b) at research level , very few theories at the time,

c) voltage / noise. to get three levels, one had to define three windows, which gave less reliable switching for the same noise, so one had to either use higher voltages or lower noise.

In the end, running logic at lower voltages , and binary reduction won,