AMC: The Platform of Choice

Aaron Schellenberg
Design Engineer
Nuvation

Due to its open architecture and high degree of flexibility, AdvancedMC (AMC) modules are quickly emerging as a key component in many carrier-grade telecommunications platforms.  Fueling the growing popularity of AMC is its interoperability, allowing manufacturers to move away from costly proprietary systems and focus on niche strengths.

Based on the AdvancedTCA (ATCA) architecture established by the PICMG consortium of telecom equipment manufacturers, the AMC.0 specification introduces modularity to the ATCA architecture by defining a framework for developing pluggable mezzanine cards on an ATCA carrier board.  The appeal of AMC is its highly flexible nature, increasing the amount of configurability, scalability, and customization in a telecom platform.

The AMC.0 specification, however, does not limit the use of an AMC module to a classic ATCA Shelf.  MicroTCA was introduced by PICMG as an additional specification for developing scaled-down, lower-cost Shelves for applications that do not warrant a full-fledged ATCA platform.  In a MicroTCA system, AMC modules are plugged directly into a backplane and become the primary workhorses of the Shelf. In keeping with the theme of flexibility, the MicroTCA specification only defines a general framework and its standard elements, leaving the actual Shelf implementation open to the manufacturer.  Further platform scalability is also possible - for example, a ‘PicoTCA’ Shelf implementation may consist of only a minimal set of MicroTCA elements and would be targeted for very small scale applications.


AMC Design Challenges

The available I/O throughout on the AMC interconnect fabric is extremely high and could exceed 100Gbps, should a designer be tempted to make use of the full capabilities allowed for in the AMC specification.  This implies that there is potential for the digital circuitry on the module itself to be very complex; certainly, there is enough I/O bandwidth to satisfy multiple NPUs, DSPs, and/or high-end FPGAs on a single module.  Yet, the realization of these devices on the module can pose significant design challenges and may be very difficult to achieve within the AMC physical constraints.

As an example, let us consider the specified AMC target PCB thickness of 1.6mm. Given this constraint, most PCB manufacturers today will be able to achieve a layer count of somewhere in the neighborhood of 16 layers using the latest materials, although this is highly design dependent.  At any rate, assuming that 16 layers is our limit, and that approximately half of the layers are dedicated power planes, this leaves us with 8 - 10 layers to route the board.  Is this enough to satisfy our requirements?

Consider some of the characteristics of a high-speed PCB: fine-pitch BGA packages with associated fan-out requirements, length matched parallel buses running between serial transceivers and central processors or FPGAs, other length matched high-speed buses interfacing to external memories, FPGAs with 1000+ I/O pins – all of this could be an incredible challenge to realize within 8 layers. 

Suppose that only 50% of the high-speed differential pairs available on the AMC finger edge are used, with each signal toggling at a rate of 2.5Gbps. These 20 diff pairs would likely be deserialized at some point on the module. At a deserialization factor of 16, this may create a total of 320 parallel data signals clocked at a speed of 125MHz or more, all of which need to be impedance controlled and length-matched to their respective buses.  Several hundred high-speed nets may not be an insurmountable challenge to route in 8 layers, but when other nets are thrown into the mix, such as external memory or inter-processor buses, the designer may start to feel a serious headache coming on.  To add to all of this are considerations such as designing for crosstalk avoidance and the minimization of the number of vias per signal, and the designer may find that the 1.6mm PCB target thickness could very well be an unrealistic goal.

Furthermore, if an AMC module were to approach the upper limit of its 80W power budget, then currents in the tens of Amps would almost surely be present on the board.  An increased amount of copper would be required to handle these high currents, resulting in 1oz, or even 2oz, planes. Thus, heavier copper means thicker power layers, introducing yet another obstacle to overcome when designing the board to meet the 1.6mm target thickness.

Additional Quicklinks:

Ask Nuvation a Technical Question
Get an Online Quote