Re: Whats it going to be
in response to
by
posted on
Mar 26, 2021 06:47AM
Good morning
Bottom is in - time to change strategy
In my previous post we have learned about « Copper cable vs optical link.
Today let’s check about lasers & modulators, a very hot topic in the industry as you see with $7 Billion Coherent Battle. And actual work POET is doing these days.
3. Laser EML vs DML including VCSEL
Overall, we can categorize optical links into two main types: (1) DML (Directly Modulated Lasers, this includes VCSEL), or (2) EML (Externally Modulated Laser).
The receiver in both methods have the same architecture where a photodiode converts optical intensity (or phase in coherent links) into electrical photocurrents. Afterwards, receiver restores the electrical currents into the digital domain via thresholding electrical circuitry.
On the transmitter side, the DML laser diode is directly modulated by applying a modulated electrical signal in DML links. On the other hand in EML approach, the laser acts as an optical continuous wavelength (CW) source and transmitter imprints digital data on the light stream via an optical modulator.
DML links require laser sources with high relaxation frequencies. However implementing such laser is very challenging and they still underlie frequency chirping. Despite the demonstrations of distributed feedback lasers (DFB) operating at 56Gb/s data-rates, DML optical links have very limited usage due to various issues. One of the major issues is that lasers cannot be monolithically integrated with CMOS driver chips and that sets an upper energy-effciency limit due to the parasitic capacitances of interconnection in between laser and electrical drive which shoule be driven at the data-rate speed. Moreover, lasers'threshold and performance depends greatly on their temperature. As using thermo-electrical cooling (TEC) is extremely power and area ineffcient, DML should operate in an uncooled mode. This also prohibit them from being co-packaged with high-performance SoCs running at 80-100 ºC normally. Finally, multi-wavelength operations like wavelength division multiplexing (WDM) links require extra optical Mux/DeMux that add extra optical path loss and area overhead.
Today DML optical links are only used for short-reach optical links via VCSEL lasers. They are multi-mode sources at 850nm leading to limited working distances of up to 300 m. Data-rates of up to around 40Gb/s have been reported with transmitter energy-effciency of 0.5-1 pJ/b. One of the advantages of VCSELs is their cost effciency, however due to challenges for achieving higher speeds and lower power consumption, their application and usage may shrink by the advent of future EML optical transceivers.
4. EML and optical modulator
EML optical interconnects bring many opportunities to improve energy-effciency and bandwidths of optical transceivers. Separating laser sources from the transmitter module solves all the issues of frequency chirping and relaxes the laser frequency metrics. Also, there's no need to drive the parasitic capacitance of laser's anode/cathodes at high-datarates and we only use DC currents to bias the laser (whether laser is 3D integrated or deployed as a separate module). Due to the temperature dependency of laser sources and also relatively low utilization of optical links in data-centers, separating laser modules from the transceiver module can eventually improve the energy-effciency and bandwidth density as the transceivers can now be placed very close to high-performance SoCs (like in packageor directly embedded on CPU/GPU die). One of the main disadvantages of EML links is the inevitable extra optical loss of coupling laser light into the transmitter chip. (POET has solved that problem)
The key device in EML links is an optical modulator that can operate based on any of the following principles: (1) Pockels effect, (2) Thermo-optics, (3) Electro-absorption, Carrier-plasma effect.
Pockels electro-optic effect, changes or produces birefringence in an optical medium induced by an electric field. However, since materials showing Pockels effect cannot be easily integrated with silicon platforms, using this type of optical modulators for silicon-photonics is still under research.
Thermo-optical effects are normally slow (<1 GHz) and cannot be used for high data-rate modulation.
Ge-based electro-absorption modulators have been demonstrated at data-rates up to 50Gb/s with relatively compact footprints. However, 100% Ge is hard to integrate into a CMOS process and the insertion loss of these devices is still large (5 dB) due to the intrinsic absorption of Ge even at lowbias voltages.
Thus electro-optical phase shifters based on carrier-plasma effect are the most promising approach for CMOS integration, which can be utilized in the either forms of a Mach-Zehnder interferometer (MZI) or a ring-resonator.
Cheers