Slashdot Mirror
Optical Control of Light on a Silicon Chip
An anonymous reader writes "Researchers at Cornell University have demonstrated a device that allows one low-powered beam of light to switch another on and off, on silicon, a key component for future "photonic" microcircuits in which light replaces electrons for propagating signals. It is highly desirable to use silicon--the dominant material in the microelectronic industry--as the platform for these photonic chips.
The approach developed confines the beam to be switched in a circular resonator, greatly reducing the footprint required on the chip and allowing a very small change in refractive index to shift the material from transparent to opaque."
Daar is nie 'n lepel nie
I thought diffraction and interference was to be the answer to switching light. Does anybody know what happened to this technology?
from the article itself.
East Coast Brewers
Light travels about 10x faster than electrons in their optimal medium, so the potential processing speed limit is increased.
Also, light processing does not necessarily generate heat, so there is no cooling needed to preserve the hardware, unlike the electro solution.
Because silicon is well established in the semiconductor industry and therefore cheap to obtain easy to process into semiconductor devices.
On the other, almost all optical devices (LEDs, laser diodes) are made from III-V compund semiconductors like Galliumarsenide (GaAs), InAs, AlAs, GaN, GaP and so on. These are not available as large crystalline blocks and thus there are no such things as 300mm wafers. They are usually fabricated by expensive methodes. However, they are the only practical solution because the are so-called direct semiconductors - you just cannot do optics with indirect band-gap semiconductors like silicon.
Now, if you find THE technological trick to do optics with silicon, you benefit from the cheap silicon technology and are ready to build optical computers with cheap fabrication technology. There are some tricks around already like mixing silicon with germanium (SiGe) or putting in nano-crystals so the silicon are catching up in doing optics.
This isn't for optical network switches, this is for processor cores.
IAAEE, so here goes a simple explanation of why optical is more desirable for a processor.
1: Faster signal propagation. In the GHz region propagation delay can cause major timing headaches in synchronous computers (one reason your system bus is always slower than your CPU: the physical length of the clock lines on the motherboard introduce too much delay to properly synchronize at really high speeds).
2: Higher slew rates. Another limit on clock speed is the rate that the logic gates can change state, which is proportional to the power consumption (it takes more power to change the state of a logic gate more quickly). Theoretically, an optical switch uses the same amount of power regardless of speed because youre switching an optical state rather than energizing (or de-energizing) a circuit.
3: Lower power consumption. Because you aren't using ever-higher currents to force electrical states at higher speeds, your driver circuitry doesn't need to be as robust. This also leads to:
4: Lower cost. Less circuitry to push around large signals means you can save die area on the chip.
This is not true, at least for this kind of optical switch. In the article, the authors state that it takes 0.15pJ to generate the free carriers. This sets a single switch to 'on', a single time, for about 500ps. If you assume that a switch is turned on, on average, 50% of the time, a single switch would consume 0.15mW. An optical CPU with one million switches would therefore need 150W, at 2 GHz. If you want a faster switch, you must reduce the carrier lifetime. Therefore you need more pump power to keep the switch turned on. So power consumption would increase linearly with clock speed.
And these numbers do not include any other losses, and assume that you can recover all the pump light which is not absorbed in the ring. If you don't recover that pump light, power consumption goes up by a factor of 166. (So you'd need 25kW for the 2GHz CPU with 10^6 switches...)
My research area is silicon based optoelectronics and we are trying to fabricate efficient light emitting silicon based components. Basic components are made from MOS-structures with incorporated excess silicon to the silicon dioxide layer. After this the device is annealed at high temperature and the excess silicon forms so called nanocrystals inside the oxide. This allows the direct electron transition like in III-V group semiconductors.
:)
In basic structures the efficiency is however very poor. All kinds of tricks are needed in order to get the efficiency in range of direct bandgap semiconductors. We do not know yet if it is possible
One of the reasons to use silicon for IC technology is its very good native oxide. You can produce dielectric with breakdown voltages of 10MV/cm with only annealing in oxygen. Think about it 100 nm of silicon dioxides breakdown voltage is over 100 V!
"Think again" is right. The electrons are involved in propagating a wave of electromagnetic energy, in ways that are fun to examine. But what you are describing is the average rate of travel of an electron, much like the average rate of travel of a lake: only a little bit of water goes in and out of it, on the average, so the average speed is very slow.
The *wave* in the lake, however, is much faster, carried by particles that bounce around each other much faster. Typical propagation speeds of electrical signals in network cable is a significant fraction of the speed of light, roughly 75% of the speed of light for 75-Ohm coax cable as one example.
Optical propagation in fiber-optic cable, which is what this new technology will be used for, is also limited to less than the speed of light. There, you get interesting effects because it's being transmitted through glass (or plastic for short cables), but still a significant fraction of the speed of light in vacuum.