Rational Ratio Multiplier (RRM) With Optimized Duty Cycle implementation

Abstract

Design and methods for implementing a Rational Ratio Multiplier (RRM) with close to 50% duty cycle. This invention gives an optimal way to implement RRM that save both area and power for a given design and able to achieve a good accuracy of the output clock with a difference between the high period and the low period of the output clock by only half a cycle of the input clock which is the closest to get to 50% duty cycle clock.

Description

FIELD

This invention relates generally to the field of Semiconductor Integrated Circuit design, and more specifically to a design implementation of Rational Ratio Multiplier (RRM).

BACKGROUND OF THE INVENTION

As Semiconductor Integrated Circuit become more complex and implementation of complex System On a Chip (SoC) are becoming widely use there are many different blocks and interfaces inside the chip which requires different and accurate clock frequencies.

Using a dedicated clock source such as crystal or PLL per each required clock frequency isn't optimal for both power and area of the silicon, thus there is a need to use one fast frequency clock source and to have a clock divider which can divide the fast clock source frequency to any required sub-frequency with good accuracy. In order to get to an accurate clock frequency this divider need to make a rational divide and for this there is a need to use Rational Ration Multiplier (RRM) which can give an accurate division and also a good duty cycle such that any logic which relies also on the falling edge of the clock can still operate.

This invention shows an optimized implementation of an RRM with close to 50% duty cycle for different giving rational ration of N/D in which D>N.

PRIOR ART

Standard 1/D Divider Implementation—FIG. 1

This simple divider will divide the clock by the 1/D ratio with a duty cycle of 50% when D is even, and with one input-clock cycle width difference when D is odd. The divider uses a counter which counts from 1 to D (box 101 in FIG. 1) and two comparators which compare the counter to D (box 102 in FIG. 1) and D/2 when D is even or D/2+1 when D is odd (box 103 in FIG. 1). When one of these two comparators finds an equal result, a toggle Flip-Flop (T-FF) (box 104 in FIG. 1) changes its state to generate the output clock. Because this T-FF is working only on the rising edge of the clock we get 50% duty cycle when D is even and one input-clock cycle difference between the high and the low of the output clock when D is odd.

1/D Divider Implementation with 50% Duty Cycle—FIG. 2

This divider will divide the clock by the 1/D ratio with a duty cycle of 50% even when the D is odd by working on both rising and falling edges of the input clock. The divider uses a counter which counts from 1 to D (box 201 in FIG. 2) and two comparators which compare the counter to D (box 202 in FIG. 2) and D/2 when D is even or D/2+1 when D is odd (box 203 in FIG. 2). When one of these two comparators finds an equal result a T-FF which works at the rising edge of the input clock (box 204 in FIG. 2) changes its state to generate the output clock in case D is even. Another T-FF which works at falling edge of the input clock (box 205 in FIG. 2) changes its state when the counter is equal D/2+1 when D is odd. A XOR (box 206 in FIG. 2) between the first T-FF which samples at the rising edge and the second T-FF which samples at the falling edge generates the output clock when D is odd. A multiplexer (box 207 in FIG. 2) selects the output of the first T-FF which samples at the rising edge or the output of the XOR depending if D is even or odd.

Because the two T-FF are working one on the rising edge of the input clock and the other on the falling edge of the input clock we get 50% duty cycle even when D is odd and not like the 1/D implementation in FIG. 1.

The disadvantage of the 1/D divider is that it can't give accurate frequency on the output as it can support only division by D of the input clock frequency and not by N/D as the RRM implementation do.

RRM Implementation Using Clock Qualifier Approach—FIG. 3

This RRM implementation creates a qualified clock pulse of the input clock at the multiplication ratio of N/D. The high pulse width of this RRM remains always one cycle of the input clock and if any logic in the design is working with the falling edge of the output clock it will be required to meet the timing constraints of the input clock which is faster than the timing constraint of the output clock.

The RRM is programmed by receiving the value of N (box 300 in FIG. 3) and D (box 301 in FIG. 3). An adder (box 303 in FIG. 3) adds at each rising edge of the input clock the value of N to the previous output of the adder which comes from the multiplexer (box 302 in FIG. 3). The multiplexer chooses between passing a zero at the beginning of the count or the stored value of a register otherwise (box 304 in FIG. 3). The register value is set according to the previous cycle's result—when the previous cycle's output of the adder is greater than D then the register will sample the result of the adder modulo (box 304 in FIG. 3). When the previous cycle's output of the adder is less than D the register will sample the adder output.

Whenever the adder result is greater than D, the clock qualifier (box 308 in FIG. 3) will receive an enable signal and pass one cycle of the input clock to the output clock. Using this method, we receive for every D cycles of the input clock N cycles on the output clock.

DESCRIPTION OF INVENTION

To get an RRM which can work on any fractional ratio and give close to 50% duty cycle a new implementation is proposed which performs this accurate clock multiplication.

This invention may be used by any system which requires an accurate clock multiplication using a rational fraction ratio and requires an output clock with close to 50% duty cycle so any negative edge logic which uses this clock can work with timing constraint of the output clock.

This invention has been described as including various operations. Many of the processes are described in their most basic form, but operations can be added to or deleted from any of the processes without departing from the scope of the invention.

RRM Implementation with Optimized Duty Cycle—FIG. 4

The RRM implementation under this invention generates an output clock with cycle time which is the multiplication of the input clock by the fraction rational ratio of N/D while D>N.

The duty cycle of the resulted output clock is close to 50% with a difference of only one half a cycle of the input-clock between the high period and the low period of output clock.

The implementation under this invention is programmed by receiving the value of N (box 400 in FIG. 4) and D (box 401 in FIG. 4). The RRM implementation calculate the low period (box 403 in FIG. 4) which is the number of rising and falling edges of the input clock pulses that the output clock should be at zero and the high period (box 406 in FIG. 4) which is the number of rising and falling edges of the input clock pulses that the output clock should be at one. The low period is calculated according to the result of the D DIV N calculation (box 402 in FIG. 4) The high period equals to the low period if N==1 (this is the same as the implementation of a 1/D divider) or equals to the low period plus one (as shown with box 404 in FIG. 4).

In addition to the above, this invention includes a counter which counts from 1 to the sum of low period+high period (box 404 in FIG. 4) using the rising edge of the input clock.

Connected to this counter are 4 T-FF's which 2 of them are sampling on the rising edge of input clock and the other two on the falling edge of input clock. These FF's are toggling when the counter equals to the following values:

• • The first T-FF (box 408 in FIG. 4) which samples at the rising edge of input clock toggles when the counter equals the high period or the sum of both the high and low periods.
  • The second T-FF (box 409 in FIG. 4) which samples at the rising edge of input clock toggles when the counter equals to half of the high period or the half of the sum of both the high and low periods.
  • The third T-FF (box 410 in FIG. 4) which samples at the falling edge of input clock toggles when the counter equals to the high period or the sum of both the high and low periods.
  • The fourth T-FF (box 411 in FIG. 4) which samples at the falling edge of input clock toggles when the counter equals to the half of the high period+1 or the half of the sum of both the high and low periods+1

When N isn't equal to 1 (this means that the implementation isn't a 1/D implementation) then the output clock is generated by a XOR (box 414 in FIG. 4) with one AND function between the first T-FF and the third T-FF (box 413 in FIG. 4) with another AND function between the second T-FF and the fourth T-FF (box 412 in FIG. 4).

When N is equal to 1 (this means that the implementation is a 1/D implementation), the output clock is generated either by a XOR of the first T-FF with the fourth T-FF (box 416 in FIG. 4) when the high period is odd, or by a XOR of the first T-FF with the second T-FF (box 415 in FIG. 4) when the high period is even. The selection between these two XOR's is done by a multiplexer (box 417 in FIG. 4) which generate the output clock if N is equal to 1.

The last multiplexer on the output (box 418 in FIG. 4) is selecting the correct clock between the above two implementation depending on whether N is equal to 1 or not.

Claims

1. A Rational Ratio Multiplier (RRM) implementation comprising: a pair of inputs of N and D which gives the rational ratio fraction N/D while D>N;two registers which store the high and low periods of the output clock counting the number of rising and falling edges of the input clock;a counter that counts from 1 to the sum of the high period plus the low period;4 T-FF's which two of them samples at the rising edge of the input clock and two at the falling edge of the input clock;random logic which generate from these 4 T-FF's the required frequency multiply output clock.

2. An RRM implementation of claim 1, wherein the high period and low period of the output clock are measured by the number of rising and falling edges of the input clock.

3. An RRM implementation of claim 1, wherein the high period of the output clock is either equal to the low period of the output clock (duty cycle of 50%) when N is equal to 1, or the high period of the output clock is larger by 1 half period of the input clock from the low period of the output clock if N isn't equal to 1.

4. An RRM implementation of claim 1, wherein the 4 T-FF's toggle as following: first T-FF which samples at the rising edge of input clock toggles when the counter equals to the high period or the sum of both the high and low periods;second T-FF which samples at the rising edge of input clock toggles when the counter equals to the half of the high period or the half of the sum of both the high and the low periods;third T-FF which samples at the falling edge of input clock toggles when the counter equals to the high period or the sum of both the high and low periods;fourth T-FF which samples at the falling edge of input clock toggles when the counter equals to the half of the high period+1 or the half of the sum of both the high and low periods+1.

5. An RRM implementation of claim 1, wherein the output clock is generated when N isn't equal to 1 by a XOR between one AND function of the first T-FF and the third T-FF with another AND function of the second T-FF and the fourth T-FF.

6. An RRM implementation of claim 1, wherein the output clock is generated when N is equal to 1 either by a XOR of the first T-FF with the fourth T-FF when the high period is odd or the XOR of the first T-FF with the second T-FF when the high period is even.