Lifting the lid on integration
Integrated circuits revolutionised the chip industry but optical ICs are creating design challenges at each new level of complexity.
By David Gahan Bookham Technology
Heated debate about integration in the optical components industry continues despite the technical advances of recent years.
The prime motivation for integration is cost. Systems suppliers also now need component suppliers to provide subsystem levels of integration so that they can then focus on areas of value-add and differentiation, rather than a basic system that merely works.
Integrated components also save space, which is a prime concern to network operators trying to squeeze extra return from their investment in facilities. Last, but not least, systems using integrated components show better reliability and lower cost of ownership.
Silicon as an integration platform
The pro-integration debate is over which technologies offer the best platforms. Initially, most companies favoured silica, the historical favourite of the components industry. Silicon, though a widely understood engineering material, had few advocates.
The dynamics have changed, however. First, the number of companies using silicon as an integration platform has grown. Second, real action finally supplemented discussion as truly integrated components reached the market.
This started with relatively simple components. The first volume-produced integrated devices based on silicon - single-fibre transceivers - began shipping in 1998.
Now deployed in hundreds of thousands of fibre-to-the-x systems, these demonstrated the potential of silicon for component integration.
The next level of silicon integration followed soon after: optical channel monitor devices combined photodetection and arrayed waveguide grating (AWG) functions; multiplexer variable optical attenuators (VOAs) integrated AWGs with solid-state, charge-injection-controlled attenuators (Fig. 1). Both are now in volume production.
The latest integration involves more complex subsystems such as optical spectrum analysis modules (OSAMs) with switching inputs. Designed for next-generation long-haul transmission systems, these are in the evaluation stage at customer sites.
Although the question of whether or not integration is a good thing is essentially answered, how integration is accomplished is still widely debated.
Many component suppliers just bundle several devices into a single package. Unfortunately this misses some of the key advantages of genuine integration. The production process for a multi-chip optical package is difficult to automate. This affects not only the cost of production but leads to inconsistencies in device characteristics whe produced in volume. Internal fibre interconnects merely replicate external ones and, without true integration, miniaturisation is restricted.
Moreover, in fibre-interconnect designs, individual packaging of key components effectively dictates the design and size of the overall package. This leads to inflexibility, both in the current design and in future development.
Which attributes determine the success of an integration platform? The four major attributes are: functionality; performance; ease of customisation; and integration yield.
Functionality is the ability of the chosen platform to perform all - or at least most of - the active and passive tasks required for optical networking. These functions can be implemented intrinsically by building structures within the material of the optical device or by hybridisation; attaching discrete devices directly to the surface of the optical substrate.
Note that this approach is not strictly monolithic. The substrate is not simply an interconnect - it must offer the ability to implement a range of functions intrinsically, with hybridisation used only where it makes sense. Unless the platform supports a range of functionality, it is not possible to produce integrated components.
Integrated components must reach, and eventually exceed, the performance benchmarks required. They must perform not just as discrete components but when combined into a fully featured subsystem. It is relatively easy to implement most optical functions in a range of technologies, but the art is to deliver the required performance when these functions are implemented in combination on the same substrate.
Customisation is required because customers increasingly ask for quick, reliable production of a number of product variants. Here the silicon approach provides a full set of optical functions capable of being slotted together on a single device, and promises rapid turnaround. Because custom solutions generally provide lower volume opportunities than off-the-shelf solutions, they traditionally fit well in the domain of hand-integration. But this approach scales poorly as volume increases, is costly and time-consuming, and is now unpopular, particularly in current market conditions.
Another key parameter is manufacturing yield. Not only are well understood, high-yield processes likely to be more commercially viable for cost reasons, components are less likely to fail in the field. The yield per function curve is an exacting one - for functional yields of less than 85-90% it is more costly to produce an integrated component than several discrete devices (see Fig. 2).
Key to functionality is the availability of a reliable switching element that can be integrated with other standard components. Customer requirements and applications vary, so a strong integration platform will allow switching to be implemented in different ways.
There are four basic factors to consider when assessing switch technologies.
- Is it a good switch? Does it perform in terms of switching speed, polarisation dependence, signal blocking etc.?
- Is it manufacturable?
- Can it be integrated?
- How much does the technology cost?
Almost any component can be regarded as a switch if it is deployed correctly and the surrounding components integrated properly. The fast switching speed and high extinction ratio of a silicon-based electronic variable optical attenuator (EVOA) is applied to very good effect in this regard as a powerful building block within many subsystem-level products.
A solid-state EVOA uses carrier injection to achieve attenuation due to free-carrier absorption within a silicon substrate. Carrier injection into the waveguide is achieved using a PIN diode structure, which allows both electrons and holes to be injected in equal proportions. This physical mechanism of attenuation means that the switching speed is about three orders of magnitude faster (about 2MHz) than thermo-optic devices. Attenuation is high (up to 60dB), enabling use for blocking applications and making the device an effective 1x1 switch.
Use of a bipolar p- and n-type structure provides efficient carrier injection - orders of magnitude greater than for unipolar devices such as resistors. The result is a device with an almost linear current/attenuation characteristic which is extremely fast, making it suitable for fast switching applications such as protection switching. It can also be produced at yields high enough to justify integration. Polarisation-dependent loss as an integrated component is virtually non-existent. Also, in contrast to some alternative implementations, it is independent of attenuation.
The EVOA is now proven to be effective in several integrated components. One is a multiplexer-VOA device that combines 80 optical functions on a single chip. This is a good example of an important design consideration when integrating, i.e. thermal management. Almost all optical materials have some thermal sensitivity. Many technologies rely on thermo-optic effects to produce useful circuits; silicon is no exception. AWGs are among the more thermally sensitive components, while current-controlled EVOAs generate heat.
This problem can be efficiently managed in a number of ways. First, silicon is intrinsically thermally conductive, so it is easy to get excess heat away from critical areas. Second, the monolithic structure enables the design of an overall package that fits in with thermal and performance requirements. Moreover, the EVOA structure is intrinsically a diode, so it can benefit from the continuous improvements in the microelectronics industry to minimise the power dissipated within the attenuators.
For higher orders of switching (2x2 and above) alternative structures can be more suitable. One implementation is Mach-Zehnder (MZ) switch, typically consisting of two 3dB couplers connected by two waveguides that serve as phase shifters (Fig. 3). When light is launched into one of the two input ports, it is split into two MZ arms by the first 3dB coupler. The two signals in the MZ arms have equal optical power and can be phase-shifted by changing the carrier concentration with a PIN diode arrangement on either arm. At the second coupler, the two beams recombine constructively or destructively at either of the two output ports, depending on the exact phase difference between the two MZ arms. The accumulated phase difference can be modulated. So, for switching light the desired output can be chosen from the two ports.
This device uses the electro-optic effects and cannot be realised in non-electro-optic materials such as silica. A similar effect can be produced in such materials by replacing the PIN diode with a thermo-optic device.
Although the couplers try to divide the optical power equally, power imbalances can occur, reducing the achievable extinction ratio. It is then possible to use MZ switches in cascade to increase the extinction ratio or deploy EVOAs to improve moderate isolation of the output signal. The cascaded MZ format also indicates a way of achieving higher-order switches in planar devices.
While switching is probably the most important function for integrated components, monitoring and control are the most critical system-level requirements for device makers.
The early single-fibre transceivers used a simple tap and monitor structure (Fig. 4) to monitor a single channel via a photodiode. After this was an integrated optical channel monitor, using AWGs and PIN arrays to isolate and monitor 40 DWDM channels.
The next step is an optical channel monitor with switched input. This integrates an 8-way EVOA-based switch, a 60-channel demultiplexer and 60 detectors into an optical processing device capable of monitoring over 400 independent channels.
Such monitoring strategies are essential for all-optical networking. Integrated components will look increasingly like optical signal processors. And investment will increase in areas such as signal analysis and software algorithms, which are essential to automate keeping the network up and running.
This is one example of why the debate over integration has been won. Customers are demanding subsystem-like optical components, and the industry must supply them. The trend can only continue as more and more companies join the debate on the side of the integrators.
David Gahan is director of DWDM products at Bookham Technology headquartered in Abingdon, Oxfordshire in the UK.