Question:
Why have CPUs been limited in frequency to around 3.5Ghz for so many years? What prevents us from increasing the frequency to 10, 100, 1000Ghz?
Answer:Oh cool I actually know the answer to a question in askscience. I am an electrical engineer and I work in the IC (integrated circuit) industry. The short answer is that scaling of semiconductor transistors has reached a limit due to the physical properties defining them. In the 1980s and 1990s, we enjoyed "ideal" scaling and were able to increase the speed and decrease the size by keeping all electric fields inside the transistor constant. As they got smaller and smaller, effects that previously could be ignored due to the macroscopic nature of the devices took a bigger and bigger role in how the devices functioned. One of the most important factors affecting constant-field scaling for devices intended to be used for digital applications (processors) is inability to scale the threshold voltage due to sub-vt operation causing leakage power. Basically the transistor is a switch, and you don't get a very good "off" if the threshold voltage is too low and every transistor inside the chip dissipates power all the time. Usually power is only dissipated in a logic circuit when it switches. Here is a good paper explaining constant field scaling.Edit: a lot of people have asked "why don't we just increase the area?" The area for chips has been roughly the same for a long time, and the reason for this is defect density and process variability. The logic behind this is that if there are a set number of defects in the manufacturing process per wafer [in #defects/cm2], then the bigger the chip is, the more area that the defect affects and you can't sell.
Read more: Reddit
QR:
Why have CPUs been limited in frequency to around 3.5Ghz for so many years? What prevents us from increasing the frequency to 10, 100, 1000Ghz?
Answer:Oh cool I actually know the answer to a question in askscience. I am an electrical engineer and I work in the IC (integrated circuit) industry. The short answer is that scaling of semiconductor transistors has reached a limit due to the physical properties defining them. In the 1980s and 1990s, we enjoyed "ideal" scaling and were able to increase the speed and decrease the size by keeping all electric fields inside the transistor constant. As they got smaller and smaller, effects that previously could be ignored due to the macroscopic nature of the devices took a bigger and bigger role in how the devices functioned. One of the most important factors affecting constant-field scaling for devices intended to be used for digital applications (processors) is inability to scale the threshold voltage due to sub-vt operation causing leakage power. Basically the transistor is a switch, and you don't get a very good "off" if the threshold voltage is too low and every transistor inside the chip dissipates power all the time. Usually power is only dissipated in a logic circuit when it switches. Here is a good paper explaining constant field scaling.Edit: a lot of people have asked "why don't we just increase the area?" The area for chips has been roughly the same for a long time, and the reason for this is defect density and process variability. The logic behind this is that if there are a set number of defects in the manufacturing process per wafer [in #defects/cm2], then the bigger the chip is, the more area that the defect affects and you can't sell.
Read more: Reddit
QR:
Posts
0 comments:
Post a Comment