Original Link: https://www.anandtech.com/show/680



Introduction

Consistency in production is a sign of a great manufacturer, one that will be with us for the duration of the ride, and not just a brief period.  Now that we have seen Intel through their good times as well as their more recent bad times, we can make a fairly accurate statement summarizing a characteristic of theirs that has not failed the company, even when the going got tough.

While there have been times when Intel hasn’t had the fastest processor on the block, as well as instances when they weren’t always the most desirable name in the industry, it can’t be refuted that Intel has always excelled in one major area: manufacturing. 

Intel’s manufacturing processes have always seemed to be one step ahead of the competition and have never failed them.  Even the latest Intel blunder, the recall of the 1.13GHz Pentium III can be attributed to a poor marketing decision to launch a chip that was obviously not 100% ready for production.  In this case, the Pentium III was limited by its architecture, the 0.18-micron manufacturing process Intel produced it on was not to blame.

Until recently, Intel hasn’t had a need to tout their manufacturing capabilities, since they’ve always been able to live off of the success of their chipsets and processors.  Now with their dominance being questioned by AMD’s astounding success Intel is just starting to get things back together again.  With the press giving Intel a very difficult time, especially with the recent Pentium 4 release, Intel is in desperate need of a bit of positive light.  Which is why today, Intel is making a fairly large announcement regarding the development of their 0.13-micron manufacturing process as well as their predictions for their upcoming 0.10 and 0.07-micron processes as well. 

What this translates to for you all is nothing much since the first 0.13-micron processors from Intel won’t hit the streets until Q2-2001 as we diagramed in our Intel CPU & Chipset Roadmap not to mention that we won’t even see 0.07-micron technology until 2005.  However this announcement does give us a unique opportunity to discuss some of what may be possible in the future, as well as shed some light on Intel’s plans for their NetBurst Architecture.



The answer lies in history

As we mentioned in our recent Pentium 4 review, there are two factors that must come into play in order for the Pentium 4 to become a truly successful chip.  The first being that the SSE2 instruction set that gives the Pentium 4 its floating-point prowess must be taken advantage of in upcoming software titles.  The second, and more pertinent to today’s discussion is that the Pentium 4 needs to ramp up in clock speed.

Believe it or not, but these are the same two characteristics that every Intel processor has required at minimum, to succeed in comparison to its predecessor.  If we make the same statement in a more general fashion we get that in order for a new processor to be successful its architecture must be taken advantage of in the current software and it must be able to increase the amount of work that it is capable of doing, quite possibly by boasting an increased clock speed.  Let’s find some examples:

Remember the introduction of the first Pentium processors?  It seemed almost silly purchasing a 5V Pentium 60 when there were cheap AMD 486 alternatives out there that didn’t require a motherboard upgrade and performed identically if not superior to these crude Pentiums.  But in the end, the 486 died off and gave way to the superior P5 architecture of the Pentium as the processor demonstrated its ability to reach amazingly high clock speeds of 200MHz and beyond. 

What about the Pentium Pro and its P6 core?  While the market was still adopting 32-bit applications, the Pentium Pro suffered as a desktop solution since it unfortunately boasted very poor 16-bit performance.  To Pentium and Pentium MMX users, the Pentium Pro’s 16-bit performance was a joke and that’s what mattered to most desktop users in 1995, remember, these users were still running DOS and Windows 3.11.  Giving the industry some time to make the move over to a predominantly 32-bit OS (Win95) Intel also took some time to play with a few of the Pro’s shortcomings and debuted the Pentium II, another P6 processor, in 1997.  The P6 quickly shed its history as an architecture to laugh at and made the Pentium II the processor of choice throughout 1998.  Again, once the architecture was taken advantage of, and the processor was given the ability to ramp up in clock speed (450MHz by the end of 1998) the CPU became a success. 

What limits clock speed

This discussion of trends in processor history is leading us somewhere.  While Intel or any other manufacturer can try to push for the architecture specific optimization in software in order to make their CPUs successful that is still largely dependent on factors that relate more to developers.  If an architecture offers no benefits to the developer or demonstrates severe drawbacks, the developing community will fail to support it. 

What CPU manufacturers can do directly to ensure the success of their processors is increase the amount of work that they are capable of doing.  This generally can be done either by increasing the number of instructions a processor can handle per clock (IPC) or by increasing clock speed.


Intel's Hyper Pipelined Technology that allows the Pentium 4 to reach higher clock speeds

The obvious question is why not increase the IPC and clock speed together, unfortunately the more work a processor is handling (the higher the IPC) the lower clocked it will have to be before running into physical limitations.  The solution to that is to decrease the IPC in favor of a higher clock speed.  This is the tradeoff both Intel and AMD have made in the past with every succeeding processor generation, the most recent example being the Pentium 4 and its hyper pipelined technology which we explain in greater detail here.

This brings us the issue of what limits clock speed.  We just discussed one limitation, the processor’s architecture, the second limitation happens to be the manufacturing process. 



Hitting switches

The transistor, invented in 1948 by Bell Laboratories is what brought about the ability for us to have these extremely complicated computers in very small packages.  The transistor is nothing more than a switch whose status (or “position” in compliance with the switch analogy) is determined by the electricity that is fed to it.  The beauty of the transistor however is that it features no moving parts, allowing it to be very small while still performing its duty.  Combining a large number of these transistors using a material that only conducts electricity under certain conditions (determined by temperature) you have what is called an integrated circuit which was first introduced just over 10 years after the invention of the transistor. 

Millions upon millions of these transistors came together in a special form of an integrated circuit, one that is not only very tiny but programmable as well.  This was the birth of the first microprocessor, the 4004 from Intel. 

The Pentium 4 and Athlon processors of today follow this same basic foundation; they are nothing more than extremely complicated descendants of the 4004.  Part of the evolutionary process has been the constant decrease of the size of these transistors.  This size is what we refer to when we say that the Pentium 4 is built on a 0.18-micron process, the 0.18-micron measurement is in reference to the size of the circuit. 

As CPUs get more and more complex their designs require more and more transistors.  However, if all other variables remain constant, we’d have processors with dies that can be measured in feet, not millimeters.  In order to keep heat production low and yields high, these transistors must get smaller as processors get faster and more complex. 


To the right you can see the actual Pentium 4 core. Kind of big isn't it?

For example, the first x86 processor, the 8088 featured 29,000 transistors with a 3-micron circuit size on a 33mm^2 die.  In comparison, the Pentium 4 features 42 million transistors with a 0.18-micron circuit size on a 217mm^2 die.  The Pentium 4’s die (surface area of the core itself) is only about 6 times as large as that of the 8080 while it has over 1400 times the number of transistors.  Can you imagine how big of a die the Pentium 4 would have if it were made on the same 3-micron process as the 8088? 



Tying it all together

Currently Intel’s processors are manufactured on a 0.18-micron process, and as we just discussed that refers to the circuit size. Next year, Intel will be shipping their first 0.13-micron parts in the first half of 2001 (we’re expecting that in the second quarter). As Intel has done for the past two generations of processors, this 0.13-micron process will debut first on their mobile processors.

The yield on any given process is lower upon the process’ introduction and increases as the manufacturing of the process matures. This is why Intel debuts all new manufacturing processes in a lower demand area (mobile processors) before moving it over to the desktop and workstation processors as well.

Before we get to what this means in terms of Intel’s future processors, let’s take a look at the actual reduction in size that the 0.13-micron process brings us. This development comes from Intel’s largest entity with 4000 workers: the Technology and Manufacturing Group (TMG).

It is from this group that we not only get the improved manufacturing processes, but we also get quite a bit of materials research from this part of Intel as well. Such as the decision to make the move to Copper interconnects with the 0.13-micron process, a decision that AMD actually beat Intel to with their 0.18-micron process.

TMG’s forecasts indicate that Intel’s manufacturing goal is for a 30% shrink in transistor size every 2 years. This results in the manufacturing forecast represented below:

Manufacturing Process Forecast
Actual
Forecast
Process Name
P852
P854
P856
P858
P860
P1262
P1264
Production
1993
1995
1997
1999
2001
2003
2005
Circuit Size
0.50µm
0.35µm
0.25µm
0.18µm
0.13µm
0.10µm
0.07µm
Gate Length
0.50µm
0.35µm
0.20µm
0.13µm
0.07µm
0.05µm
0.03µm

To illustrate exactly how small a the 0.07-micron process is, Intel uses the comparison: transistors as small as DNA. While that statement does have its flaws, let's have a look at a picture of this comparison first before extracting the truth from it.


John Jackson & Inman. Gene 1989 84 221-226.

Here we have a picture of a 0.01-micron (10nm) Gold particle attached to a Z-DNA antibody.

And now we have a transistor with a gate length of 30nm or 0.03-microns, just three times as the Gold particle attached to the Z-DNA antibody from the picture above.



Does the 0.13-micron Pentium 4 spell success?

As we discussed in our Intel Roadmap preview last month Intel plans to release a new version of their Pentium 4 core, codenamed Northwood, in Q4-2001 at 2GHz and beyond. 

Northwood will be a 0.13-micron version of the Pentium 4 that will boast a die that is approximately 1/2 the size of the 217mm^2 Pentium 4 (Willamette) core.  This will not only increase yield on the Pentium 4, but it will make it much cheaper to manufacturer as you can produce many more chips per silicon wafer which currently measure 8” in diameter (P8xxx process).  After the Northwood debuts, Intel should be debuting their 12” wafers (P12xx process) allowing even more processors to be produced per wafer, decreasing production costs even more as time goes on.

The Northwood will run at 1.30v compared to the 1.70v core voltage of the current Willamette based Pentium 4, and will boast significantly reduced operating temperatures as well.  For example, a 2GHz Northwood may produce less heat than a 1.5GHz Pentium 4. 

This opens the avenue for the Pentium 4, and the NetBurst architecture in general to be seen on some very high clocked processors.  How high you may ask? 



10GHz by 2005 running at < 1 volt

The P6 micro-architecture was introduced with the Pentium Pro at 150MHz in 1995 and is still with us today with the Pentium III at 1GHz.  The P6 architecture will be with us for a little while longer, in the end offering clock speeds close to 1.3GHz if not higher which is close to a 9x increase in clock speed since the architecture’s introduction.  Thus it isn’t too far fetched to assume a similar scalability from the Pentium 4’s NetBurst architecture. 

Realistically speaking, we should be able to see NetBurst based processors reach somewhere between 8 – 10GHz in the next five years before the architecture is replaced yet again.  Reaching 2GHz isn’t much of a milestone, however reaching 8 – 10GHz begins to make things much more exciting than they are today.  Obviously this 8 – 10GHz clock range would be based on Intel’s 0.07-micron process that is forecasted to debut in 2005.  These processors will run at less than 1 volt, 0.85v being the current estimate.

This brings up the obvious question, what would you possibly want to do with a 10GHz processor?

Running Microsoft Word can only take so much processing power, regardless of how complex your documents may be, so there’s no real need for such a powerful processor in conventional application areas.  However there are areas that definitely could use a little more power. 

Intel is stressing their focus on pattern recognition as it applies to voice dictation and face detection among other things. 

Imagine being able to speak normally with your computer as you would a secretary sitting next to you and have your computer accurately and quickly take notes from your speech. 

Imagine logging onto your computer not via a user name and a password but by sitting in front of your display and having it scan your face to figure out if you are allowed access to the computer.   

These are things that Intel is claiming will be possible by 2005 with the type of processors that will be available in desktop systems.  It’s definitely going to be a bright future if this does hold true.  Intel is working very hard in developing the software that will help make these visions a reality.  Just think of all that has happened in the past 5 years alone, is the future that we’re painting for you all that far fetched of an idea? 



Facts from Intel

In closing, we’d like to leave you with a set of 9 interesting facts Intel supplied us with regarding the processors and their transistors we will be intimately familiar with by 2005.  Enjoy:

1.      The transistors are 0.03 microns wide and some of their structures are about 3 atomic layers thick.  The transistors are so small that a vertical pile of 30 million transistor gates would measure 1 inch high (12 million for a centimeter) and a stack of more than 100,000 would equal the thickness of a sheet of paper.

2.  These Transistors can turn on and off 10 billion times per second

3. These transistors will be built into Intel processors that are nearly 10 times more complex than the Intel® Pentium® 4 processor, today's most advanced processor.  For example, the future processors will have 400 million or more transistors, will run at 10 GHz and operate at less than 1 volt. Today’s Pentium 4 processor has 42 million transistors, runs at 1.5 GHz and operates at 1.7 volts.

4. Faster than a speeding bullet: A 10GHz processor will be able to complete 20 million calculations in the time a speeding bullet travels 1 foot, or 2 million calculations in the time it travels 1 inch

5.      A 10GHz processor is faster than the blink of an eye.  In the times it takes you to blink (1/50th of a second or so), the processor can complete about 400 million calculations

6.      Imagine putting 400 million parts on a chip the size of your finger nail.  Rio de Janeiro’s Maracana stadium, one of the world’s largest athletic stadiums, could only contain an estimated 200,000 spectators for the largest crowd ever gathered for a football (soccer) game.

7.      At 1 volt or less, these future processors will consume significantly less power than today’s processors.  Thus, they could be easily used in battery-operated devices such as laptop computers and handheld devices.  As a comparison, today’s Pentium 4 processor operates at  1.7 volts.  (AnandTech Note: obviously this one is stretching it a bit since we know that voltage isn’t the only thing that matters when it comes to portable devices)

8.      These transistors will begin appearing in products manufactured using 0.07-micron (70 nanometer) technology, which is three manufacturing processes more advanced than Intel's current 0.18 micron technology.  To put this in perspective, today's 0.18-micron technology is two manufacturing processes more advanced that the 0.35 micron technology used when the Pentium processor was introduced in 1993.  The Pentium® processor had 3.1 million transistors, ran at 66 MHz and operated at 5 volts. 

9. The 0.07 micron (70 nanometer) technology will rely on Extreme Ultra Violet (EUV) lithography, the next generation lithography technology, for printing the narrowest lines.  This will be combined with 157nm lithography to enable manufacturers to continue producing smaller and faster processors.  Lithography is the process in which circuits -- the pathway through which electrical current flows -- are printed on silicon wafers.  EUV will allow semiconductor manufacturers to print ever-smaller features on a wafer.  The difference between features drawn by EUV and Deep Ultra-Violet (DUV) lithography, today’s most advanced method, is similar to drawing two lines of equal width and quality on a piece of paper, but using a fat-tipped marker to draw one line and a fine-tipped marker for the other.

Log in

Don't have an account? Sign up now