Recommended Design Practices

• Transmission lines. EMI often becomes issues when you try to get the FCC certification for your product. Transmission lines dramatically reduce EMI, because transmission lines are constructed to keep returning signal current close to the outgoing signal path. The magnetic fields from the two currents cancel each other. The typical transmission lines on PCB are microstrip (traces on the top and bottom layers) and stripline (traces on the inner layers).

• Terminations. To minimize the reflection of signals on the board, terminations are recommended. Source terminations are normally used in CMOS and TTL logic design, because they do not require high sink/source current. CMOS and TTL drivers usually only can sink/source maximum 24mA current. End terminations are normally used in the ECL and PECL logic design. SSTL2 logic (e.g. DDR interface) requires both source and end terminations.

• Clock distribution. Clock toggles faster than other signals, and connects every flip-flop on the board. It has to be distributed carefully to ensure the system work properly. To minimize the reflection, every clock line should employs a source terminator. The value of the termination resistor depends on the number of clock lines and trace impedance. The length of the each clock line should be the same t minimize the clock skew. Thus it requires the trace length matching in the layout.

• Power distribution. Power plane is highly recommended to minimize the voltage drop across the board. Ground plane is highly recommended to minimize the common-path noise voltages. Place bypass capacitors close to the power supplies output pins for board level filtering, and also place bypass capacitors close to the power pins of each chip for local filtering. Use low ESR bypass capacitors to minimize the power voltage ripples.

• PCI and PCIe bus. The clock trace length should be less than 2" from the card edge to the chip input pin. The length of each AD bus line should be the same. The differential pair of the RX, TX and clock of PCIe should be length-matched.

• Board stacking. The way of stacking the PCB board is very important for impedance control. Below is a typical 62mil thick 10-layer PCB stacking (6 layers signals and 4 layers power/ground) with uniform impedance on both micro strip lines and stripelines. 

• Design with a synchronous clock. The logic clouds need a synchronous clock to optimize the timing closure. Make sure the main clock is fan-outted by a global clock buffer. Avoid to use local clocks in the design. 

• Design with a synchronous reset. An asynchronous reset from external world can be spurious resets due to noise or glitches on the board or system reset. If these spurious reset pulses occur near a clock edge, the flip-flops can go metastable. A reset synchronizer is recommended to be employed. A reset synchronizer can be a pair of cascaded flip-flops reset by the asynchronous reset and synchronized by the global clock. The synchronized reset drives through the a reset global buffer to the rest of the flip-flops in the design. 

• Pipeline stages. To achieve the desired high-speed logic design, pipeline stages are recommended to add. One example would be the realization of FIR filter. A number of MAC (Multiply-Accumulate) operations needs to be used. The data computing path needs to be broken down into multiple pipeline stages to achieve the timing. 

• Offset constraints. One the circuit board often the FPGA interfaced with external devices such as memory chips, CPU bus, other FPGAs etc. To successfully interface with the external devices, the setup and hold time has to be met. Specifying offset constraints in the constrain file will ensure to meet the setup and hold time when FPGA talks to the external devices. 

• I/O pads. There are flip-flops and delay elements inside I/O pads. Generally-speaking you want to ensure the FPGA input and output signals are registered inside the I/O pads for achieving the best timing. You may also intentionally turn on/off the delay elements to adjust the timing. You may examine the place and route results inside the I/O pads by using the vendor provided tools such as Xilinx FPGA editor.

• Multi Pass Place and Route. Sometimes the single pass place and route does not create the best possible timing. You may chose multiple pass place and route option by specifying the number of passes and the cost table value, which save the best place and route results on the computer.