doubling, then doubling again, then doubling again is x^2, not 2^x. Your original example described geometric growth. Exponential growth would be: 2, 16, 65536, (very large number)
what you described was: 2, 4, 8, 16, 32, 64,...
This is geometric growth. It is a common mistake to label things exponential growth. True exponential growth is exceedingly rare, and is unsustainable.
of the most impressive technological feats our society accomplishes on a regular basis. The45nm process has nothing to do with innovation. It's just the same technology, the same process, on a different scale.
What you declare is simply not true.
45nm is the result of a huge amount of innovation, just as 65nm was compared to 90nm. There are a lot of technological hurdles to overcome as the length of transistors are scaled. For example, improved high-k dielectrics are required to increase the channel capacitance and reduce leakage. Improved isolation between devices is required. Tighter tolerances for lithography are needed. Better control of ion implant doses are required. More stable silicides are needed to reduce interconnect resistance. Better drain structures are needed to deal with the increased electric field density in the transistor channels. Improved thermally conductive materials need to be developed because the heat density is increasing. I could go on and on and on. Scaling transistors is onere is a huge financial incentive to do so, and tens of thousands of engineers worldwide are attacking the problems from many angles.
What most people don't understand about device scaling is that it isn't a single problem to be solved. It is a huge number of equally challenging problems spanning multiple engineering disciplines.
I'd be hard pressed to believe anyone is releasing professionally pressed CDs from master recordings that clip.
I wish that were true, but sadly it is not. I 100% agree with you the a recording will not sound worse after peak normalizing to 0 dB. The issue is what consumers notice is not the peak normalization but the perceived RMS volume. If people hear two CDs and one is louder, most people will indicate the louder CD is better. In order to increase the RMS volume, master engineers have been compressing the audio, then cranking up the gain and even gently clipping it. This is why newer CDs are so loud at the same volume level compared to CDs from, say, the late 80s.
You can see this for yourself. If you have a digital audio editing software, such as Audiology, you can see the time-domain waveform. Take the latest Depeche Mode CD, "Playing the Angel", and compare the waveform to "Violator" and you will see what I mean. Also listen to the tonal differences of the recording. Besides the loudness difference, you will notice a harshness to the newer recording. A friend of mine looked at the waveform from the record of Playing the Angel and found it did not clip. He claims that it sounds better than the CD for this reason. I have not heard the record so I cannot comment.
No one has pointed this out, but early CDs actually sounded like shit. They were WAY under compressed so the noise of the signal path was very significant. Also, they were mostly encoded at the studio in 16 bit, so mulitiplies and stuff going on in the mastering process (most were mastered on analog equipment but there was a required digital transfer to the master). The early CDs I heard in the mid 80s were really trashy. Better than tapes, but not as good as records on a decent hi-fi. This was pretty much common knowledge amongst people who liked music. The real selling point, and what made me get on eventually, was the random track access. That was huge, and I believe that is really what made the CD take off.
Interestingly, there was a kind of golden-era of CD sound in the late 90s when we had high dynamic range mastering equipment, before the loudness war pissed it all away in a hail of clipping.
The joke I heard 20 years ago was some guy was trying to get a job with AT&T but they found out his parents were married when he was born and wouldn't hire him.
Wansu,
I honestly don't understand the joke. Could you please explain it?
Almost as many chips are built by third parties today as built internally. Intel is the only American company still developing leading edge CMOS right now. THere are a lot of semiconductor design companies that outsource fabrication to Taiwan. Some of the biggest are Broadcom, Xilinx, Altera, Marvell.... It makes good business sense, since the huge captial and NRE costs of fabrication are spread over a lot of customers, lowering the price for all.... the big downside is that it makes it harder to differentiate in commodity markets since the fabricators are similar. That is why commodity chips like RAMS and processors tend to be built in-house currently. The third-party foundries (in particular TSMC and Chartered) are working to move into commodity businesses.
As for the 65nm fab... it's not the wafers that kills... it's the masks. Third-party fabrication won't help with that!
IANAPL, but patents based on improvements on other patents still require licensing from the original patent holder to produce. If you could simply bypass a patent by making a slight improvement, then I'm sure Blackberry and RIM suits would have never happened. This is simply not true. The vast majority of patents are improvements on previously disclosed inventions. The original patents are listed as references. The key is the improvement has to be non-obvious to someone skilled in the art. This is where the lawyers come in. In the case of infringement, it is almost always someone didn't know about the patent or, more likely, tried to get away copying it. If there is a true improvement, then a new patent can be issued.
That's true, and actually with current silicon device sizes a single alpha particle strike has the possibility of flipping a bit in an SRAM. This is one part of why NASA uses old cpus -- one of the simplest methods of radiation hardening is to simply use larger structures that require a larger amount of energy to change state. Then they add more shielding and such on top of course.
That's actually not true at all. The chance a transient error (SRAM bit flip) or worse, a long term change in the threshold voltage of a device actually gets worse when the structures are larger. That is because the chance for a radiation event to occur in the gate oxide is linearly proportional to the thickness of the oxide. Fine-line CMOS has thinner oxides, so it is more tolerant.
On top of that, what you are discussing (shielding, structure geometry) is called radiation tolerance, not radiation hardening. A radiation hard IC process implies dielectric isolation between the devices. For example, the use of SOI is quite prevelent in nuclear/space applications. The reason NASA uses old CPUs is because they are available in rad-hard dielectrially isolated technology. Intersil in Palm Bay, FL, still has rad-hard 286s coming off the line right now. Dielectrically isolated IC processes with the feature sizes needed to produce modern CPUs simply do not exist because of the lack of an economic incentive. That is the only reason NASA and DOD use such old CPUs.
Quite honestly, I would view your military background as more favorable than a BS. My experience with ex-military sysadmins/operators has been excellent. I would still recommend that you go for a BS degree part-time if you were working for me, but any lack of one would be no problem. You would look far more valuable to an enlightened employer than a new grad. Good luck.
That is so interesting that you come from a completely different place. An entire Atari 2600 game would only be a couple of hundred lines of code, but each line was important, and usually somewhat deep. It's kind of like that phase "being the machine". I just loved the way you could almost "feel" how the 6502 worked when you programmed in assembly. Many things were just not possible in Basic, Pascal or C. Nowadays it's the opposite. I imagine only a true expert could write a better x86 game in ASM than in C. I certainly couldn't. I bet I would do best in Python, which is the closest I can get to the kind of "fun" I remember in the early 1980s. The joke in college, though, when people would argue between C, Lisp, and Turbo Pascal as their favorite programming language, I would always reply "My favorite programming language is solder!"
I read that as "2600 game developer awards", and was looking forward to a fun nostalgic look at 6502 assembly programming. Oh well. I guess I can also look at the Atariage archives.
I get so nostalgic about programming. It was just so much fun in the early 80s. I work in IC design now, and the programming is primarily Perl, Python, and proprietary scripting languages, but I really miss BASIC, Pascal, and Assembly on those early machines.
I'm a bit confused by your reasoning, spun. A few posts ago you were saying that Quantum Computing would someday enable us to model the Quantum state of each brain molecule, then you're saying NO quantum mechnical processes play a part in consciousness. First off, the complexity of quantum entanglement grows geometrically with the number of quanta in question. It is analgous to calculating the quantum probability density functions (PDFs) of electrons in an element. It's easy to do for Hydrogen, harder for He, then quickly becomes practially impossible. I expect it will be the same in the brain. While in some Platonic fantasy of Penrose's, maybe you could calculate, but it would take an infinite amount of time on an infinitely powerful computer.
And, for the record, a strong argument for the choatic (actually it is properly called Stochastic) nature of the brain is the following: Signals are transmitted between neurons via action potentials (electro/chemical reactions). These action potentials can be quite well modeled as filtered Guassian random processes (that is the underlying process is random, but the action potentials are correlated in time, so they aren't truly random). Now, computing a PDF of one of them is tough enough, but then the joint PDF of billions is IMPOSSIBLE.
The brain is pretty much an f-ed up analog computer, and we are further from understanding conciousness today than we were 10 years ago.. if you catch my drift. I imagine your Ph.D. student friends are aware of all this.
That really is how "they" name things. For example semiconductors: SSI - small scale integration MSI - medium scale integration LSI - large scale integration VLSI - very large scale integration ULSI - ultra large scale integration
The same holds true for complex numbers just as well as for integers. The problem comes with embedding those truths in our physical universe. I can hand you four apples. You can't hand me 3i+1 oranges in return.
But I can... you just have to define the imaginary part in a different way, for example heigth or something. The magnitude would be the number of oranges (3 oranges plus a couple of sections in your example), the phase of the number would somehow be related to how tall I am. That is exactly how modern radio receivers work, two sine waves, offset in phase, one is assigned a real number, the other an imaginary number, so both can be manipulated simultaneously as a complex number.
Normally, two floating point values are used to represent complex arithmetic, however its not a native operation, and still requires some software logic to be accomplished.
It was always my understanding that humans don't know what a complex number is either. They are not fundamental in the sense integers or real numbers are... they are a mathematical convenience (an invention). They can be used to describe amplitude and phase in the same breath, or Fourier components, but neither application requires complex arithmetic. It's just taking the one dimensional real number line and adding a second, orthogonal one. A complex number is just a two dimensional number, so the way a computer deals with it is no different from the way a human would have to deal with it.
side from some very important differences between patents and copyright (patents have a non-renewable short lifetime and are not transferable)
Actually, patents are quite transferable. I have two, both transferred to my employer. For $1 each. I get nothing if they are used. It is part of the employment contract. At least the company did pay for the notary to sign my rights away!
Total myth. bg knows what lines of code look like on a piece of paper, and that's pretty much it. He hasn't ever coded anything to functional completion.
Not true. BG wrote a BASIC interpreter for the Altair, and some other stuff too. He is a very smart man, and knows computers. Don't underestimate him or the technical prowess of the engineers at Microsoft.
Slashdot Mirror
doubling, then doubling again, then doubling again is x^2, not 2^x. Your original example described geometric growth. Exponential growth would be: 2, 16, 65536, (very large number)
...
what you described was:
2, 4, 8, 16, 32, 64,
This is geometric growth. It is a common mistake to label things exponential growth. True exponential growth is exceedingly rare, and is unsustainable.
of the most impressive technological feats our society accomplishes on a regular basis. The45nm process has nothing to do with innovation. It's just the same technology, the same process, on a different scale.
What you declare is simply not true.
45nm is the result of a huge amount of innovation, just as 65nm was compared to 90nm. There are a lot of technological hurdles to overcome as the length of transistors are scaled. For example, improved high-k dielectrics are required to increase the channel capacitance and reduce leakage. Improved isolation between devices is required. Tighter tolerances for lithography are needed. Better control of ion implant doses are required. More stable silicides are needed to reduce interconnect resistance. Better drain structures are needed to deal with the increased electric field density in the transistor channels. Improved thermally conductive materials need to be developed because the heat density is increasing. I could go on and on and on. Scaling transistors is onere is a huge financial incentive to do so, and tens of thousands of engineers worldwide are attacking the problems from many angles.
What most people don't understand about device scaling is that it isn't a single problem to be solved. It is a huge number of equally challenging problems spanning multiple engineering disciplines.
You start with no business and the next year, you have twice as much, and so on.
That isn't exponential growth... that is geometric growth.
I'd be hard pressed to believe anyone is releasing professionally pressed CDs from master recordings that clip.
I wish that were true, but sadly it is not. I 100% agree with you the a recording will not sound worse after peak normalizing to 0 dB. The issue is what consumers notice is not the peak normalization but the perceived RMS volume. If people hear two CDs and one is louder, most people will indicate the louder CD is better. In order to increase the RMS volume, master engineers have been compressing the audio, then cranking up the gain and even gently clipping it. This is why newer CDs are so loud at the same volume level compared to CDs from, say, the late 80s.
You can see this for yourself. If you have a digital audio editing software, such as Audiology, you can see the time-domain waveform. Take the latest Depeche Mode CD, "Playing the Angel", and compare the waveform to "Violator" and you will see what I mean. Also listen to the tonal differences of the recording. Besides the loudness difference, you will notice a harshness to the newer recording. A friend of mine looked at the waveform from the record of Playing the Angel and found it did not clip. He claims that it sounds better than the CD for this reason. I have not heard the record so I cannot comment.
No one has pointed this out, but early CDs actually sounded like shit. They were WAY under compressed so the noise of the signal path was very significant. Also, they were mostly encoded at the studio in 16 bit, so mulitiplies and stuff going on in the mastering process (most were mastered on analog equipment but there was a required digital transfer to the master). The early CDs I heard in the mid 80s were really trashy. Better than tapes, but not as good as records on a decent hi-fi. This was pretty much common knowledge amongst people who liked music. The real selling point, and what made me get on eventually, was the random track access. That was huge, and I believe that is really what made the CD take off.
Interestingly, there was a kind of golden-era of CD sound in the late 90s when we had high dynamic range mastering equipment, before the loudness war pissed it all away in a hail of clipping.
My cousin's bones are sitting somewhere in North Vietnam after being shot down while flying an F-4... you insensitive clod!
There actually was an innovative microprocessor called "Clipper". It was a nice architecture...
http://en.wikipedia.org/wiki/Clipper_architecture
It isn't a conservative idea: conservatism eschews getting unnecessarily embroiled in costly occupations, especially in Asia.
Especially land wars in central Asia.
The joke I heard 20 years ago was some guy was trying to get a job with AT&T but they found out his parents were married when he was born and wouldn't hire him.
Wansu,
I honestly don't understand the joke. Could you please explain it?
Carl
IBM is punting on pure CMOS after 65nm. I don't have the reference, but that is my understanding.
Almost as many chips are built by third parties today as built internally. Intel is the only American company still developing leading edge CMOS right now. THere are a lot of semiconductor design companies that outsource fabrication to Taiwan. Some of the biggest are Broadcom, Xilinx, Altera, Marvell.... It makes good business sense, since the huge captial and NRE costs of fabrication are spread over a lot of customers, lowering the price for all.... the big downside is that it makes it harder to differentiate in commodity markets since the fabricators are similar. That is why commodity chips like RAMS and processors tend to be built in-house currently. The third-party foundries (in particular TSMC and Chartered) are working to move into commodity businesses.
As for the 65nm fab... it's not the wafers that kills... it's the masks. Third-party fabrication won't help with that!
IANAPL, but patents based on improvements on other patents still require licensing from the original patent holder to produce. If you could simply bypass a patent by making a slight improvement, then I'm sure Blackberry and RIM suits would have never happened.
This is simply not true. The vast majority of patents are improvements on previously disclosed inventions. The original patents are listed as references. The key is the improvement has to be non-obvious to someone skilled in the art. This is where the lawyers come in. In the case of infringement, it is almost always someone didn't know about the patent or, more likely, tried to get away copying it. If there is a true improvement, then a new patent can be issued.
That's true, and actually with current silicon device sizes a single alpha particle strike has the possibility of flipping a bit in an SRAM. This is one part of why NASA uses old cpus -- one of the simplest methods of radiation hardening is to simply use larger structures that require a larger amount of energy to change state. Then they add more shielding and such on top of course.
That's actually not true at all. The chance a transient error (SRAM bit flip) or worse, a long term change in the threshold voltage of a device actually gets worse when the structures are larger. That is because the chance for a radiation event to occur in the gate oxide is linearly proportional to the thickness of the oxide. Fine-line CMOS has thinner oxides, so it is more tolerant.
On top of that, what you are discussing (shielding, structure geometry) is called radiation tolerance, not radiation hardening. A radiation hard IC process implies dielectric isolation between the devices. For example, the use of SOI is quite prevelent in nuclear/space applications. The reason NASA uses old CPUs is because they are available in rad-hard dielectrially isolated technology. Intersil in Palm Bay, FL, still has rad-hard 286s coming off the line right now. Dielectrically isolated IC processes with the feature sizes needed to produce modern CPUs simply do not exist because of the lack of an economic incentive. That is the only reason NASA and DOD use such old CPUs.
jascat,
Quite honestly, I would view your military background as more favorable than a BS. My experience with ex-military sysadmins/operators has been excellent. I would still recommend that you go for a BS degree part-time if you were working for me, but any lack of one would be no problem. You would look far more valuable to an enlightened employer than a new grad. Good luck.
That is so interesting that you come from a completely different place. An entire Atari 2600 game would only be a couple of hundred lines of code, but each line was important, and usually somewhat deep. It's kind of like that phase "being the machine". I just loved the way you could almost "feel" how the 6502 worked when you programmed in assembly. Many things were just not possible in Basic, Pascal or C. Nowadays it's the opposite. I imagine only a true expert could write a better x86 game in ASM than in C. I certainly couldn't. I bet I would do best in Python, which is the closest I can get to the kind of "fun" I remember in the early 1980s. The joke in college, though, when people would argue between C, Lisp, and Turbo Pascal as their favorite programming language, I would always reply "My favorite programming language is solder!"
I read that as "2600 game developer awards", and was looking forward to a fun nostalgic look at 6502 assembly programming. Oh well. I guess I can also look at the Atariage archives.
I get so nostalgic about programming. It was just so much fun in the early 80s. I work in IC design now, and the programming is primarily Perl, Python, and proprietary scripting languages, but I really miss BASIC, Pascal, and Assembly on those early machines.
I'm pretty sure you've been trolled.
I'm a bit confused by your reasoning, spun. A few posts ago you were saying that Quantum Computing would someday enable us to model the Quantum state of each brain molecule, then you're saying NO quantum mechnical processes play a part in consciousness. First off, the complexity of quantum entanglement grows geometrically with the number of quanta in question. It is analgous to calculating the quantum probability density functions (PDFs) of electrons in an element. It's easy to do for Hydrogen, harder for He, then quickly becomes practially impossible. I expect it will be the same in the brain. While in some Platonic fantasy of Penrose's, maybe you could calculate, but it would take an infinite amount of time on an infinitely powerful computer.
And, for the record, a strong argument for the choatic (actually it is properly called Stochastic) nature of the brain is the following: Signals are transmitted between neurons via action potentials (electro/chemical reactions). These action potentials can be quite well modeled as filtered Guassian random processes (that is the underlying process is random, but the action potentials are correlated in time, so they aren't truly random). Now, computing a PDF of one of them is tough enough, but then the joint PDF of billions is IMPOSSIBLE.
The brain is pretty much an f-ed up analog computer, and we are further from understanding conciousness today than we were 10 years ago.. if you catch my drift. I imagine your Ph.D. student friends are aware of all this.
Carl
That really is how "they" name things. For example semiconductors:
SSI - small scale integration
MSI - medium scale integration
LSI - large scale integration
VLSI - very large scale integration
ULSI - ultra large scale integration
what's next?
The same holds true for complex numbers just as well as for integers. The problem comes with embedding those truths in our physical universe. I can hand you four apples. You can't hand me 3i+1 oranges in return.
But I can... you just have to define the imaginary part in a different way, for example heigth or something. The magnitude would be the number of oranges (3 oranges plus a couple of sections in your example), the phase of the number would somehow be related to how tall I am. That is exactly how modern radio receivers work, two sine waves, offset in phase, one is assigned a real number, the other an imaginary number, so both can be manipulated simultaneously as a complex number.
Normally, two floating point values are used to represent complex arithmetic, however its not a native operation, and still requires some software logic to be accomplished.
It was always my understanding that humans don't know what a complex number is either. They are not fundamental in the sense integers or real numbers are... they are a mathematical convenience (an invention). They can be used to describe amplitude and phase in the same breath, or Fourier components, but neither application requires complex arithmetic. It's just taking the one dimensional real number line and adding a second, orthogonal one. A complex number is just a two dimensional number, so the way a computer deals with it is no different from the way a human would have to deal with it.
Actually, patents are quite transferable. I have two, both transferred to my employer. For $1 each. I get nothing if they are used. It is part of the employment contract. At least the company did pay for the notary to sign my rights away!
what the hell is 1T-SRAM? How can it be SRAM without feedback?!?!
Sorry mate, cat's dead.
Total myth. bg knows what lines of code look like on a piece of paper, and that's pretty much it. He hasn't ever coded anything to functional completion.
Not true. BG wrote a BASIC interpreter for the Altair, and some other stuff too. He is a very smart man, and knows computers. Don't underestimate him or the technical prowess of the engineers at Microsoft.