Sei sulla pagina 1di 4


All Digital Phase Locked Loop Design and Implementation

Anitha Babu, Bhavya Daya, Banu Nagasundaram, Nivetha Veluchamy University of Florida, Gainesville, FL, 32608, USA

AbstractAn all digital phase locked loop was implemented, in 0.25 micron CMOS technology, by understanding the analog phase locked loop concepts and the digital conversion required to maintain the same functionality. The all digital phase locked loop achieves locking within about 100 reference clock cycles. The pure digital phase locked loop is attractive because it is less sensitive to noise and operating conditions than its analog counterpart. The implementation and comparison of the digital and analog phase locked loops are performed in this paper.

Index TermsDigital PLL, DPLL, Phase locked loop, PLL, Phase frequency detector, DCO



Many circuits currently face the problem of clock skew, and registers and flip-flops are not receiving the clock at the exact same time. The clocks are generated by oscillators, but the clocks that reach the registers and flip-flops are distorted and require a phase locked loop to address this problem. A phase locked loop ensures that the clock frequencies seen at the clock inputs of various registers and flip-flops match the frequency generated by the oscillator. The phase locked loop (PLL) is a very important and common part of high performance microprocessors. Traditionally, a PLL is made to function as an analog building block, but integrating an analog PLL on a digital chip is difficult. Analog PLLs are also more susceptible to noise and process variations. Digital PLLs allow a faster lock time to be achieved and are attractive for clock generation on high performance microprocessors.

for clock generation on high performance microprocessors. Figure 1: High Level PLL System Diagram The all

Figure 1: High Level PLL System Diagram

The all digital phase locked loop was designed such that it is composed of four main components. The components are

analogous to the analog PLL, but the implementation consists of digital components. A digitally controlled oscillator (DCO) was utilized instead of a voltage controlled oscillator. A high level block diagram of the implementation is shown in Figure 1. 0.25 micron technology was used for the implementation of this project. The output frequency of the PLL is four times the input frequency. The supply voltage of the system is 2.5 V. The Phase Frequency Detector (PFD) detects the phase and frequency mismatch of the reference clock and divided DCO clock. The PLL is locked when the PFD detects that the phase and frequency of the two clock inputs match. The output of the PFD drives the time to digital (T2D) converter. The PFD produces up and down enable signals that are interfaced to the T2D converter. The T2D converter takes these inputs and increases or decreases the control word which is fed to the thermometric decoder. This decoder is essential for controlling the DCO. The DCO clock is divided by a specific multiplication factor, in our case it is four, and sent back to the PFD for phase and frequency comparison. The main components and their implementation will be discussed in the following sections. The implementation of the entire PLL doesn’t contain any VHDL based components. All the components were created schematically and the layout was completed by hand.


A. Digital Phase/Frequency Detector

The phase frequency detector is a significant aspect of the PLL because it determines whether the reference clock and divided DCO clock are in phase and are running at the same frequency. The design used is shown in Figure 2.

and are running at the same frequency. The design used is shown in Figure 2. Figure

Figure 2: Phase Frequency Detector Diagram


A modified D Flip-flop was utilized because the D input doesn’t change and remains high always. The output of the modified D Flip-flops enters a two input NOR gate that resets the Flip-flops if both clocks are high. The up and down signals indicate if the DCO clock needs to be increased (up is true) or decreased (down is true). The event and direction signal are necessary to create the up and down enable signals for the T2D converter. Additional circuitry between the PFD and T2D is required for the signal conversion to take place.

and T2D is required for the signal conversion to take place. Figure 3: Output of PFD

Figure 3: Output of PFD when Refclk and DCO divided Clk phase difference varies.

As can be seen in Figure 3, when the phase of the DCO divided clock varies by a larger margin, the pulse width of the up signal increases with respect to the phase error. The up signal contains the phase error because the DCO clock is falling behind and needs to be increased in frequency to achieve locking. The design and understanding of the PFD was important when continuing with the design and implementation of the other components of the system.

B. Time to Digital Converter

The time to digital converter consists of a 6 bit down counter, 6 bit up counter, and 6 bit carry ripple adder. The phase detector controls the up counter and down counter by up and down enable signals. The initial state of the up counter is “000000” and the down counter is “111111.” The up counter and down counter values are input into the six bit adder and the output produces the seven bit control word for the DCO. Figure 4 shows the connections of the T2D converter. The six bits from the adder and the carry out bit compose the seven bit control word. The converter should be active only if there is a phase and/or frequency mismatch. Clock gating has been performed to disable the T2D when both the reference clock and divided DCO clock are locked. The outputs from the carry ripple adder emerge at different times. This caused the decoder to interpret its inputs erroneously. In order to correct this error, a register made of D flipflops was used to store the outputs from the adder. The seven bit control word was released from this register based on

the clock given to the counter, delayed by an amount equal to the propagation delay of the counter and the adder. This method of synchronizing the adder outputs did not work as planned and was eliminated from the final PLL system design.

Figure 4: Time to Digital Converter Diagram

system design. Figure 4: Time to Digital Converter Diagram C. Thermometric Decoder The thermometric decoder is

C. Thermometric Decoder

The thermometric decoder is a specific decoder that generates the digital word that controls the DCO. The thermometric decoder was designed by modifying the general decoder to obtain the required functionality. The decoder generates a 128 bit output, of which only 126 bits are used by the DCO. These 126 bits control the DCO’s tristate gates.

D. Digitally Controlled Oscillator

A standard digitally controlled oscillator was implemented, which consists of a seven stage ring oscillator (in 0.25 micron technology with an inverter delay close to 20ps and the frequency of 1.4 GHz) and 21 tristates are parallel to each inverter. The ring oscillator contains a NAND gate that is used to control the DCO. This DCO is easy to make using standard cells, but its power consumption is quite high. In [2], it was mentioned that fifty percent of the power consumption of a PLL comes from the DCO. Figure 5: DCO Consisting of Ring Oscillator and 21 Tristate inverters per inverter. One of the main design criteria for a DCO is to provide

One of the main design criteria for a DCO is to provide enough control word resolution

enough control word resolution and maintain acceptable jitter. [2] If the resolution is kept small, the jitter of the PLL can be maintained. Jitter is an important digital PLL performance


parameter that is simulated for this system. Our DCO design was tested for all the seven bit input sequences using a linear feedback shift register. The frequency corresponding to each control word was noted. The operating range of the DCO is 180 MHz to 320 MHz. For a control word of “0000000” the DCO frequency is 48.93 MHz and for “1111111” it is 322.9 MHz. The DCO resolution is close to 1 MHz. Figure 6 displays for each control word the frequency of the DCO.

6 displays for each control word the frequency of the DCO. Figure 6: Graph of DCO

Figure 6: Graph of DCO Analysis

An optimized DCO was developed to be analyzed with regard to power, but due to time limitations this was not performed. The optimized DCO should have consumed less power and resulted in a more energy-efficient PLL.

E. Frequency Divider

The output of the DCO needs to be divided to match the reference frequency. The implementation contains a divide by four frequency divider, enabling the PLL to have a multiplication factor of four. The divider is implemented by using two series D Flip-flops. The DCO clock is input into both flip-flops and the non-inverted output of the second flip- flop is the DCO clock divided by four.


The final system is a consolidation of all these components and extra buffers and circuitry. These main components form an autonomous system that finds the DCO frequency that is four times the reference frequency. The final system runs until the lock is obtained and the T2D is no longer active because

the lock is obtained and the T2D is no longer active because lock occurred. The PFD

lock occurred. The PFD continues to analyze the divided DCO clock signal with the reference clock signal for any variations. A frequency or phase variation will cause the PFD to become activated and the synchronization of two clocks will be achieved again. Figure 7 shows the schematic of the PLL system, a better resolution image is attached.

Figure 7: System in Cadence with all Components


The PLL was simulated and found to lock at approximately

1.5 us, when the reference frequency was equal to 70 MHz.

This means the lock occurs in about 105 clock cycles. When the all digital PLL is compared to an analog PLL datasheet, from Jameco website, the lock time of an all digital PLL was found to be much smaller than the 10 us lock time of

an analog PLL. The normalized phase noise for most PLLs with integrated VCOs was found to be -212 dBc/Hz

On this website, the maximum supply voltage is shown to be

3.2 V. The cost of the analog PLL is approximately $6.65 each

when thousand ICs are purchased. The output power of an analog PLL to a 50 ohm load is about 3.5 dBm.

power of an analog PLL to a 50 ohm load is about 3.5 dBm. Figure 8:

Figure 8: PLL Output and Locking Occurs The jitter measurement of the PLL is performed by using Cadence to determine the jitter of the DCO output. The jitter analysis diagram is shown in Figure 9.


4 Figure 9: Jitter Analysis The area, power and delay of the system, determines the cost

Figure 9: Jitter Analysis

The area, power and delay of the system, determines the cost of the phase locked loop. The comparison of the digital PLL to the analog PLL is shown in Table 1. The area of the layout is 45666.5 um^2. The power of used by the PLL system is 3.875 mW, which is equivalent to 5.883 dBm. The number of transistors in the digital PLL system is 3194.

Table 1: Comparison of Analog and Digital PLL


Digital PLL

Analog PLL


5.883 dBm

3.5 dBm

Lock time



10 us

(100-120 cycles)

Jitter/Phase Noise



-212 dBc/Hz

The lock time of the digital PLL was observed to be a great improvement over the analog PLL. The power of the digital PLL is more than an analog PLL possibly because of the DCO being used. If a lower power DCO was implemented into the system, the power could decrease significantly.


The modified D Flip-flops used for the PFD has lower power consumption. The clock gating used in the T2D also saves power. Other low power implementations can be lowering the power consumed by the DCO, by using mode selection. Multiple supply voltages can be used, by having the lower Vdd supply to the counter, adder, and DCO.


The VHDL implementation was started in parallel with the schematic approach. Since the schematic approach of the PLL required us to perform layout by hand, the VHDL all digital PLL implementation was left aside. If this was completed, an analysis could have been performed on the power, area, and lock time differences of the two implementation approaches. The thermometric decoder can be made faster; at this point it is a large and slow decoder that occupies a large part of the layout. If the decoder can be optimized, the system should be smaller and more efficient. The phase maintenance can be used to load the final control word for the DCO from a register instead of continually performing phase comparison.



The development of an all digital phase locked loop was interesting and a great learning experience. There are many improvements and concepts that still need to be learned, but the basics of an all digital PLL implementation was covered during this project. A good lock time was achieved of 105 cycles and considerations for power were implemented in this project.


We would like to thank the TA, Tanuj Aggarwal, for his constant support with debugging simulation errors and logical errors. Thank you to Professor Bashirullah for making this learning experience possible.



“A Digitally Controlled PLL for SoC Applications” Thomas Olsson, and


Peter Nilsson IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 39, NO. 5, MAY 2004 751 “An All-Digital Phase-Locked Loop with 50-Cycle Lock Time Suitable for High-Performance Microprocessors” Jim Dunning,, Gerald Garcia, Jim Lundberg, and Ed Nuckolls, IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 30, NO. 4, APRIL 1995


“A fully integrated standard-cell digital PLL,” T. Olsson and P. Nilsson, IEEElectron. Lett., vol.37, pp. 211212, Feb. 2001.

[4] J.M. Rabaey, A. Chandrakasan, and B. Nikolic., “Digital Integrated Circuits”, 2nd ed.Prentice Hall, 2003, ISBN 0-13-120764-4. [5] N. Waste and K. Eshraghian, “Principles of CMOS VLSI Design”, Addison-Wesley, 1993.