I recently came across the TDK SESUB-PAN-T2541 Bluetooth module. It is a complete bluetooth module consisting of a Texas Instruments CC2541 Bluetooth “System-on-Chip” (SoC) and the peripheral components required to use it – crystal, balun, and 15 other passive components. While it initially doesn’t sound that remarkable, when you look at the size (4.6mm x 5.6mm) and the thickness (1mm) you realize that it is remarkably small. It achieves this feat by embedding the die of a CC2541, which would normally be 6x6mm and 1mm thick when packaged in the PCB. It got me thinking about what other techniques there are around to minimize circuit size, particularly with the current need to make smaller and smaller portable products.
Embedding simple passive components, particularly capacitors, is one way of reducing overall size. Of the passives in the TDK T2541, the majority are decoupling with most of the rest being tuning components. While they are conventional surface mount in this case, another strategy would be to use a conventional IC and embed some passive components. Obviously. the TDK module doesn’t take this approach because the packaged CC2541 is larger than their overall PCB.
The surface area of a PCB is premium space and while you can have more and more inner layers as well as blind and buried vias, the increase in surface area created by those efforts is small. Three-dimensional component construction seems a great idea in theory.
A product I am currently working on has a 144-pin FPGA which requires 34 decoupling capacitors. Depending on the FPGA package and decoupling capacitor package chosen, the decoupling capacitors can take up more space than the FPGA itself. Murata GRU series and AVX UT series capacitors are both intended for embedding within PCB substrates and have copper termination rather than tin in order to facilitate embedding. Both the Murata capacitors and the AVX ones are 0.15mm thick and 1mm x 0.5mm overall. The Murata web site gives an illustration of the area saved by embedding the capacitors:
In reality, you could put decoupling capacitors on the opposite side of the PCB to the chip but in a system with limited space the underside is probably already in use. AVX has a short paper on embedded components although it is uncertain if the document is current as it is undated (but appears to be from 2011). While embedding capacitors is easier than embedding die, if you wanted to go down the route of embedding decoupling capacitors, what is involved?
You need to choose your components based on the manufacturing technique e.g. solder versus copper connections, and you need a PCB manufacturer who can work with the embedded components as they need to be embedded during the PCB fabrication. To avoid problems when soldering the PCB surface components, the embedded components are often chosen with copper electrodes and connected with copper vias. This prevents potential failures from melting solder when soldering the top/bottom components, which could happen if solder was used to connect the embedded components Finding a PCB manufacturer to also place the embedded components will be your first challenge – most PCB manufacturers aren’t set up to place components. Then you need to design the PCB in a way which allows the manufacturing to be carried out, such as creating the inner cavities required. Altium have added support for embedded PCB components in their PCB layout package, allowing the cavities to be defined for active or passive embedded components, for example.
If your PCB layout package doesn’t support embedded devices you will still be able to design the PCB, it will just require more intervention to create manufacturing data which contains the required information, and accompanying notes and discussions with the PCB manufacturer.
An alternative approach if you are simply looking to add embedded capacitance for decoupling would be to use a PCB material and construction that creates additional power plane capacitance using a thin, high dielectric material for one or more of the PCB layers. 3M have such a material, as do other material manufacturers such as Sanmina, DuPont, and Mitsui. The 3M system uses a 19µm dielectric layer with a dielectric constant of 21 for embedding between the power plane layers instead of FR4 (or other PCB material). It has around 6nF per square inch of capacitance. Other presentations from 3M suggest a dielectric as thin as 6µm with 20nF/in≤ and it seems there is the possibility of materials with dielectric constants in the thousands being available. An Oak-Mitsui presentation provides useful information here:
While 6nF/in≤ doesn’t really seem to provide enough capacitance to decouple a typical FPGA, it could eliminate most of the decoupling capacitors, leaving just a few larger ones which can be placed conventionally. This is the implication from Sanmina which suggests a large reduction in the number of capacitors required with their BC 2000 dielectric core from 48 capacitors to 17 for a 1156 BGA. Presumably, the remaining conventional capacitors are the higher values.
As a result of looking into PCB embedded components I came to the conclusion that while the technology exists and has done for quite a while, it is still rather niche with components and fabricators hard to find. There will also undoubtedly be a cost penalty, so you need to be really certain that is the only way to reduce you system size before you use it, unless your product isn’t price-sensitive.