Create your own embedded Linux computer

Create your own embedded Linux computer

All of today's best integrated circuits are welded on a large scale using BGA encapsulation. Because the BGA encapsulation method is connected under the chip and the welding is tighter, the Reflow Soldering box or hot printing plate must be used. Another problem is that when designing a PCB (Printed Circuit Board), the welding ball between the pass hole and the lead should be small enough, and more layers are usually needed on the motherboard to free up space for the next lead, this means that a cheap Chinese-made two-tier motherboard does not have enough space, so more layers are needed. The additional layer will significantly increase the cost of the motherboard, even if there are only a few more copies.

I want to design a motherboard with a built-in BGA chip to experience how difficult it is to weld them. So I decided to design a small Linux-running ARM embedded system. The ARM processor used was AT91SAM9N12 in a 217-ball LFBGA package, it is only because it is the cheapest among ARM processors with memory management units necessary to run Linux. At first I only wanted to use a BGA chip, but the RAM in the BGA Package was much cheaper than other packages, so I decided to add a DDR2 (Double Data Rate 2) memory in the BGA Package.

Locate holes to maximize available space.

As a result, it was quite a week-long to find the motherboard manufacturer. Two layers of space are not enough. At least four layers are required. The diameter of the ball in the 217-LFBGA package is 0.4mm, and the distance from the ball to the face is 0.8mm. In order to leave more space for the passing hole, the layout of the ball pad is slightly smaller than that of the ball. I used 36mm mm pad. Placing holes between four balls maximizes the available space. The manufacturer must be able to manufacture a hole that can be placed in the 8mm width. Almost all manufacturers can make this diameter, but the problem is: this distance includes the diameter of the drilled hole, twice the width of the bypass, twice the minimum distance between the pass hole and the lead. For example, for the iTead layer-4 main board, the minimum diameter of the drilled hole is 0.3mm, the minimum ring width is 0.15mm, and the minimum distance between the drilled hole and the lead is 0.15mm, which adds up to 0.9mm, this means that the minimum size of the passing hole cannot be placed between the bgaball. The only vendor I found that can meet this requirement and the price is relatively reasonable. Their 4-layer motherboard has a smaller limit, and the passing hole can just be placed into the bgaball. The additional benefit is that it is cheaper for a small motherboard than iTead.

The smallest pass-through under the OSH park design principle can just be put down.

Even if the passing hole can be placed in the middle of the BGA ball, there are still some problems: there is not enough space to strip in the middle of the passing hole. This means that it is impossible for each pad to have a standard cabling channel through holes. This means that the motherboard needs to have enough unconnected pad, so the cables need to be arranged from the inside. Fortunately, the processor also has many general I/O pins that are not connected.

... If the design principle is not violated, the overlay will not be able to pass between two overpasses. CAS does not have enough space for passing through DQM0 and D15.

After the production problem is solved, it is time to think about which parts should be placed on the motherboard. I don't really care about the practical use of this motherboard. Compared with the actual use, the entire project is a learning process. The size of the motherboard is smaller to reduce costs. This means that no additional interface space is reserved, such as Ethernet, serial port, or SD card.

In addition to the processor and RAM, other necessary components are: large memory, voltage regulator, and Monitoring Circuit for Chip reset. The processor can be started from NAND, but in case I decide to add Dataflash (Data flash) to the boot loader, although it will be rarely used in the end. For large memory, NAND is a good choice because of its large capacity and low cost. Adding a BGA Package will be cheaper, but I have been overwhelmed by two BGA packages, so I decided on a 48-pin TSOP (thin and small size package) the package uses 4 gb nand. Connecting each component has been well explained in the processor list, but it is still difficult to find all the details in the documents on the last thousand pages. Atmel also released a schematic diagram of the trial board, which will be helpful when designing the motherboard.

DDR2 lead space should have a certain degree of freedom. The proper lead length should be appropriate, with controllable impedance and can terminate or concatenate resistance. In the reference design of the Development Board, all DDR2 signals use series resistors. I don't have enough space for them, so I decided to leave them alone for the moment. The impedance is not 50 ohm, because I have to use smaller leads to fill other spaces. I hope that, because RAM is closer to the processor, even if the series resistance box is missing or the impedance does not match, it does not matter. All connections from CPU to RAM are about 25mm long. The general experience is that if the lead length is more than 10% of the signal wavelength, the impact of the conversion line should be taken into account. This means that the frequency is approximately 1 GHz or above. The clock frequency of RAM is only 133 MHz, and the first several harmonic waves are still below 1 GHz, which indicates that it should work properly. To ensure the feasibility, I almost completely matched the length of the lead, but this may not be necessary.

For more details, please continue to read the highlights on the next page:

1 2 3 4 5 Next Page