EETimes: ST plans for Dresden FDSOI production

[Idontcare]

But getting back to the noobish part, how can the logic shrink in area when the interconnects larger than the xtors? I suppose there's allot more 'empty' space at a microscopic level than I'm aware off, but that's all I've got.

Planar transistors are 2D animals, and those same 2 dimensions are in play with 2D finfets as well.

Now marketing likes to focus our attention on the sexiest dimension - the length of the transistor - because it is measured in tens of nanometers and everyone likes talking about how tiny their bits are

The xtor length is what is shown to you in all those sexy SEM/TEM images, the distance between the source and the drain, i.e. they show you the channel.

Now there is a good technical reason why engineers focus on the xtor length - this is the one that determines important parameters such as the switching speed.

But the other dimension, called the transistor width, is the one that doesn't get much attention in marketing but is the one that carries the bulk of the burden of shunting enough current across the channel to power the circuits at a given targeted clockspeed.

Idrive is reported in terms of a normalized parameter. Say 100 nA/um. The "per micron" is telling you the transistor width necessary to hit that 100nA value.

It is this transistor width that scales with Idrive. And this is true with FinFets as well.

So if GloFo increases the Idrive of their xtors with 14XM then it is very possible that chips designed for 20nm could experience a shrink in terms of not requiring the metal wiring to span as large of a transistor width as they need for the 20nm designs with its weaker transistors.

 

[Phynaz]

IDC, Could we have one of your awesome illustrations please?

 

[Blandge]

IDC, Could we have one of your awesome illustrations please?

Below is a classic CMOS cross section diagram which consists of two transistors. The length of the channel is the length of the substrate between the n+ (for the NMOS) and p+ (for the PMOS) regions. This is the number given by manufacturers to refer to a specific process technology (45, 32nm, 28nm, 22nm, 20nm). For the most part this number is fixed as it defines the entire circuits default characteristics. The length of the channel is determined by the process engineers (chemical engineers, physicists, mathamaticians).

The width of the channel can differ for each transistor in the circuit depending on the chracteristics on the adjecent devices. This is one of the major components of circuit design and layout. Below is a top down view of the diagram shown above. The length is shown by the width (the small dimension) of the input shape, and the width is shown by the height of the n+ and p+ diffusion regions. It's not rare to find a transistor with a width that is 10-20x the length of the channel (or more) if it is driving an interconnect or large circuit block. The reason why the process is defined by the length is because that's what gives the circuit it's baline characteristics, but the logic is actually designed by manipulating the width (among other things).

Disclaimer: I don't do circuit design or layout as a profession.

 

[Phynaz]

The length is shown by the width

Cleanup in isle 5 - my head just exploded :biggrin:

 

[Blandge]

Cleanup in isle 5 - my head just exploded :biggrin:

The reason it's refered to as length (what intuitively we would think of it as width) is because it's the length that the holes must travel to diffuse across the channel from drain to source in order to provide current to the device. It makes sense when you consider the width is the number of holes that can diffuse in parellel, across the channel. Think latency (length) the time it takes for the holes to diffuse across the channel, vs bandwidth (width) the amount of holes you can pack into the channel side-by-side.

In both diagrams the current flows right to left or left to right. (From the inside to the outside of the picture)

 

[pm]

Cleanup in isle 5 - my head just exploded :biggrin:

Yeah, I remember thinking that same thing back in my VLSI class in college.

It's like when they told me that electrons actually travel from negative to positive, not positive to negative.

Or that there is a charge carrier called a hole which is the lack of an electron. "Wait, you are saying that charge is carried by the non-existence of an electron? How the heck does that work?"

BTW, nice explanation, Blandge.

 

[Blandge]

Yeah, I remember thinking that same thing back in my VLSI class in college.

It's like when they told me that electrons actually travel from negative to positive, not positive to negative.

Or that there is a charge carrier called a hole which is the lack of an electron. "Wait, you are saying that charge is carried by the non-existence of an electron? How the heck does that work?"

BTW, nice explanation, Blandge.

Your assumption is correct. In electrical engineering we refer to holes so that you're can think of current and charge carriers flowing from in the same direction from positive to negative. The labeling convention of positive and negative was erroneously standardized by a scientist who didn't know enough about electromagnetism to make this decision, and because of it, we live in a world where current flows in the opposite direction as the charge carriers. That's why the idea of "holes" was created which means exactly what you said, the lack of electrons flowing from positive to negative.

 

[Abwx]

It is either holes or carriers and they have different diffusion speed ,
hence the N channel fets are systematicaly faster than their P channel
siblings that will ultimately set the max frequency of a circuit.

 

[Ajay]

Thanks IDK and Blandge :thumbsup:

 

[Blandge]

It is either holes or carriers and they have different diffusion speed ,
hence the N channel fets are systematicaly faster than their P channel
siblings that will ultimately set the max frequency of a circuit.

From Wikipedia:

• In semiconductors (the material used to make electronic components like transistors and integrated circuits), in addition to electrons, the travelling vacancies in the valence-band electron population (called "holes"), act as mobile positive charges and are treated as charge carriers. Electrons and holes are the charge carriers in semiconductors.

Here's a good breakdown http://www.ioffe.rssi.ru/SVA/NSM/Semicond/Si/electric.html

Although it's all kind of meaningless unless you're using it to implement something.

 

[Abwx]

Either holes or electrons , if you prefer , but neverless
positive and negative charges have different velocities
and this will translate in an assymetry in max frequency.

In short , dynamicaly speaking , nfets are almost a node
further than pfets.

 

[ehume]

The labeling convention of positive and negative was erroneously standardized by a scientist who didn't know enough about electromagnetism to make this decision

The scientist in question was Benjamin Franklin. Yes, THE Benjamin Franklin. We have had plenty of opportunity in the subsequent centuries to "fix" the "problem."

 

[Idontcare]

The scientist in question was Benjamin Franklin. Yes, THE Benjamin Franklin. We have had plenty of opportunity in the subsequent centuries to "fix" the "problem."

Too bad they didn't go with the metric system either :\

 

[ehume]

Too bad they didn't go with the metric system either :\

Dr. Franklin discovered that electricity flows in currents. it is interesting that no one has disturbed his +/- convention to indicate flow.

The metric system, of course, was not imposed on France until the Empire of Napoleon, some time after Franklin's death.

Personally, for personal distance I like inches and feet. For distances less than an inch I think in English and metric, and under 1/8 inch I think in metric. That's just my instinct. But I think the two can cohabit nicely, as long as engineers don't work in metric. Then we will have probes crashing into Mars . . .

 

[ShintaiDK]

Too bad they didn't go with the metric system either :\

Just because of thìs?

Well, maybe in 50 years you might finally be on metric

 

[ehume]

I won't argue the distances. The English system is person-centered and sentimental for that reason. So is the month-day--year system. But I will say that I find the day-month-year system, while logical, inconvenient when you expand from two-digit years to four-digit years. For dates, the four-digit year, followed by month and day works perfectly. And you can even append the time: 2013-02-15 1603 PST.

 

[Ajay]

Just because of thìs?

Well, maybe in 50 years you might finally be on metric

Ugh, Please don't remind me. After Carter said we were switching I was taught in metric. Then we didn't change over. Thank God that I've worked as a technician, scientist and engineer (or been a student of the same) a good part of my adult life :thumbsup:

 

[Idontcare]

Then we will have probes crashing into Mars . . .

LOL, so true, and so sad.

Just because of thìs?

Well, maybe in 50 years you might finally be on metric

Exactly.

I won't argue the distances. The English system is person-centered and sentimental for that reason. So is the month-day--year system. But I will say that I find the day-month-year system, while logical, inconvenient when you expand from two-digit years to four-digit years. For dates, the four-digit year, followed by month and day works perfectly. And you can even append the time: 2013-02-15 1603 PST.

That is what I have evolved to adopt as well, year-month-day. It is the only system that numerically sorts correctly when rank-sorting. I didn't care about date nomenclature until my livelihood came to depend on my programing, and my programing depends on revision history, which depends on effective ranking of one's records and log entries...which critically depends on have a system that actually can be numerically rank-sorted.

Year first, then month, then day - it is the only way to correctly capture chronology when sorting.

 

[Ajay]

That is what I have evolved to adopt as well, year-month-day. It is the only system that numerically sorts correctly when rank-sorting. I didn't care about date nomenclature until my livelihood came to depend on my programing, and my programing depends on revision history, which depends on effective ranking of one's records and log entries...which critically depends on have a system that actually can be numerically rank-sorted.

Year first, then month, then day - it is the only way to correctly capture chronology when sorting.

That's pretty trivial programmatically (in C++, Java, etc.), or do you mean using other tools like Excel + macros/VBA?

 

[Idontcare]

That's pretty trivial programmatically (in C++, Java, etc.), or do you mean using other tools like Excel + macros/VBA?

I mean log records that are created with timestamps. Programmatically speaking, one could argue everything is trivial. In practice I found it was not so until I went to year-month-date format.

Maybe I am just weaksauce in this regard, really it would not surprise me as mostly all of what I know of programming has been self-taught.

 

[Abwx]

Year-month-day is falsely intuitive.

Seconds - minutes - hours - days - months - years.

That s the most logical since we start from the durations
that have the least significant absolute values on the left to the most
significant absolute values on the right , just as the counting system ,
be it binary or decimal.

 

[ehume]

Year-month-day is not intuitive at all. But as IDC notes, it sorts well within apps -- no programming needed.

 

[nehalem256]

Year-month-day is falsely intuitive.

Seconds - minutes - hours - days - months - years.

So why do all digital clocks go

Hour: minute: second then?

That s the most logical since we start from the durations
that have the least significant absolute values on the left to the most
significant absolute values on the right , just as the counting system ,
be it binary or decimal.

Actually I think you have it backward.

Biggest numbers go on the left.

ie 1000 > 0001

so likewise years are bigger than days so they belong on the left

 

[Ajay]

I mean log records that are created with timestamps. Programmatically speaking, one could argue everything is trivial. In practice I found it was not so until I went to year-month-date format.

Maybe I am just weaksauce in this regard, really it would not surprise me as mostly all of what I know of programming has been self-taught.

Doesn't mean you are weaksauce, it's just not the same as programming professionally 50-60 hours a week for years. It'd be like someone handing me a couple of books on semiconductor process development and asking me to do your job (actually, that would be a disaster D.

 

[Idontcare]

Doesn't mean you are weaksauce, it's just not the same as programming professionally 50-60 hours a week for years. It'd be like someone handing me a couple of books on semiconductor process development and asking me to do your job (actually, that would be a disaster D.

LOL, me doing my job was a disaster Thankfully my employer didn't see it that way until all the checks were cashed. Me and Ruiz, twins separated at birth