The idea of fault tolerable computing is similar to the EnerJ programming language being developed at the University of Washington for power savings The Language of Good Enough Computing
The jist of the idea is that the programmer can specify which variables need to be exact and which variables can be approximate. The approximate variables would then be stored a low refresh RAM which is more prone to errors to save power, while the precise variables would be stored a higher power memory which would be error free.
The example they gave was calculating the average shade of grey in a large image of 1000 by 1000 pixels. The running total could be held in an approximate variable since the error incurred by adding one pixel incorrectly out of a million would be small, while the control loop variable would be accurate since you wouldn't want your loop to overflow.
It looks like the ISPs are standing up to traffic throttling in in response to Google, Amazon and Skype's request to the CRTC to ban traffic throttling.
With big recognizable names like Google, Amazon and Skype backing net neutrality, hopefully the CRTC will be swayed to rule in favor of stopping traffic shaping or at least scrutinize the current behavior of ISPs like Rogers and Bell.
Google, Amazon ask CRTC to stop Internet traffic shaping
Chipworks based in Ottawa Canada specializes in chip reverse engineering for patent litigation and technical analysis. They have reverse engineering reports ranging from the Xbox360 to CPUs to analog chips. They also have a neat chip art gallery.
From a technical standpoint I would agree that fiber is far more superior, but from an economic standpoint it does not make much sense. The consumer electronic market is extremely price sensitive. Adding optical components will add expensive and unnecessary costs to the final product, which could be avoided by just using a piece of copper as the cabling and living with its impairments.
At minimum you would need to add an optical transmitter like an LED or LASER, and a photodetector at the receiver. Both of these components are relatively expensive as they are made exotic III-IV materials. This means they they can't be integrated with the rest of the CMOS display driver electronics, and must reside as a separate chip. The more chips that need to be put on a board the more expensive the product becomes.
Optical cabling would make sense in high-end products where the consumer is willing to pay extra for it, but for the masses, the lower the price the better.
What they did sounds like an extension of the technique used in push-push oscillators to "double" the oscillation frequency.
The basic principle behind a push-push oscillator is that two out-of-phase signals of fundamental frequency f_o are combined such that the fundamental signal and the odd harmonics cancel, while the second harmonic at 2*f_o add constructively. In the case of a push-push oscillator, you only need two signals 180 degrees out of phase. This could be generated with a differential VCO.
Using a push-push oscillator is a well known technique for increasing the frequency of oscillation of a VCO beyond the fMAX of a transistors at a given process node.
The only disadvantage with push-push oscillators is that you end up losing a lot of power as the second harmonics's power will always be much smaller than the power in the fundamental frequency of the VCO.
It seems what the MIT scientists have created is a purely optical equalizer in a convention CMOS process. This would probably be used at the receiving end of a single mode fiber link. Most of the equalization done today is done electronically using fancy optical receivers (expensive but very robust).
The article is light on details but the idea of integrating photonics and electronics in a conventional CMOS process isn't a new idea. Maybe the way they did the integration is a breakthrough. A company called Luxtera demonstrated (with products) integrated photonic and electronic transmitters way back in 2005. Their press release from March 2005 http://www.luxtera.com/news_press_2005_0328.htm reveals that they created an optical modulator (a transmitter) in Freescale's CMOS process. The optical modulator they created is also based on the same idea of splitting light and combining it to create on/off pulses at extremely high speeds.
There is still a clock in the optical domain. The bits are passed along the optical channel via on and off light pulses which exist for a certain amount of time. So a "1" is the presence of light and off is no light at all. This similar to how a clocked computer uses electric on and off pulses where "1" is a high voltage and "0" is a low voltage. Without going into too much detail about clock and data recovery circuits, the duration that these pulses stay on and off can be thought of as clock period.
By using optical links, this breakthough will solve some of the problems we have today with sending data at high speed across chip to chip busses. The major problem today with sending data at high rates between chips is the losses incurred by travelling across the FR-4 PCB. As the data rates go up, the greater the losses incurred, the more difficult it is to recover the data being sent. Optical interconnects have significantly less losses at high data rates, thus making them a suitable technology for chip to chip communications in the future. This is a breakthrough because now we can integrate exotic optical materials with low cost silicon using standard chip-making equipment. This was something that could not be done in the past.
Slashdot Mirror
The idea of fault tolerable computing is similar to the EnerJ programming language being developed at the University of Washington for power savings The Language of Good Enough Computing
The jist of the idea is that the programmer can specify which variables need to be exact and which variables can be approximate. The approximate variables would then be stored a low refresh RAM which is more prone to errors to save power, while the precise variables would be stored a higher power memory which would be error free.
The example they gave was calculating the average shade of grey in a large image of 1000 by 1000 pixels. The running total could be held in an approximate variable since the error incurred by adding one pixel incorrectly out of a million would be small, while the control loop variable would be accurate since you wouldn't want your loop to overflow.
Posting to undo accidental mod.
It looks like the ISPs are standing up to traffic throttling in in response to Google, Amazon and Skype's request to the CRTC to ban traffic throttling.
With big recognizable names like Google, Amazon and Skype backing net neutrality, hopefully the CRTC will be swayed to rule in favor of stopping traffic shaping or at least scrutinize the current behavior of ISPs like Rogers and Bell.
Google, Amazon ask CRTC to stop Internet traffic shaping
Use the lower case "b" (AbCdEF)
Chipworks based in Ottawa Canada specializes in chip reverse engineering for patent litigation and technical analysis. They have reverse engineering reports ranging from the Xbox360 to CPUs to analog chips.
They also have a neat chip art gallery.
From a technical standpoint I would agree that fiber is far more superior, but from an economic standpoint it does not make much sense. The consumer electronic market is extremely price sensitive. Adding optical components will add expensive and unnecessary costs to the final product, which could be avoided by just using a piece of copper as the cabling and living with its impairments.
At minimum you would need to add an optical transmitter like an LED or LASER, and a photodetector at the receiver. Both of these components are relatively expensive as they are made exotic III-IV materials. This means they they can't be integrated with the rest of the CMOS display driver electronics, and must reside as a separate chip. The more chips that need to be put on a board the more expensive the product becomes.
Optical cabling would make sense in high-end products where the consumer is willing to pay extra for it, but for the masses, the lower the price the better.
What they did sounds like an extension of the technique used in push-push oscillators to "double" the oscillation frequency.
The basic principle behind a push-push oscillator is that two out-of-phase signals of fundamental frequency f_o are combined such that the fundamental signal and the odd harmonics cancel, while the second harmonic at 2*f_o add constructively. In the case of a push-push oscillator, you only need two signals 180 degrees out of phase. This could be generated with a differential VCO.
Using a push-push oscillator is a well known technique for increasing the frequency of oscillation of a VCO beyond the fMAX of a transistors at a given process node.
The only disadvantage with push-push oscillators is that you end up losing a lot of power as the second harmonics's power will always be much smaller than the power in the fundamental frequency of the VCO.
It seems what the MIT scientists have created is a purely optical equalizer in a convention CMOS process. This would probably be used at the receiving end of a single mode fiber link. Most of the equalization done today is done electronically using fancy optical receivers (expensive but very robust).
The article is light on details but the idea of integrating photonics and electronics in a conventional CMOS process isn't a new idea. Maybe the way they did the integration is a breakthrough. A company called Luxtera demonstrated (with products) integrated photonic and electronic transmitters way back in 2005. Their press release from March 2005 http://www.luxtera.com/news_press_2005_0328.htm reveals that they created an optical modulator (a transmitter) in Freescale's CMOS process. The optical modulator they created is also based on the same idea of splitting light and combining it to create on/off pulses at extremely high speeds.
If you want to read more about their technology and why integrating photonics with electronics is important visit: http://www.luxtera.com/technology_faq.htm
There is still a clock in the optical domain. The bits are passed along the optical channel via on and off light pulses which exist for a certain amount of time. So a "1" is the presence of light and off is no light at all. This similar to how a clocked computer uses electric on and off pulses where "1" is a high voltage and "0" is a low voltage. Without going into too much detail about clock and data recovery circuits, the duration that these pulses stay on and off can be thought of as clock period.
By using optical links, this breakthough will solve some of the problems we have today with sending data at high speed across chip to chip busses. The major problem today with sending data at high rates between chips is the losses incurred by travelling across the FR-4 PCB. As the data rates go up, the greater the losses incurred, the more difficult it is to recover the data being sent. Optical interconnects have significantly less losses at high data rates, thus making them a suitable technology for chip to chip communications in the future. This is a breakthrough because now we can integrate exotic optical materials with low cost silicon using standard chip-making equipment. This was something that could not be done in the past.