I. INTRODUCTION
As the data rate increases in serial interfaces, a higher electromagnetic interference
(EMI) level is becoming an issue for system design. There are various approaches to
reduce EMI levels such as shielding, slew-rate control, low voltage differential clocking,
pulse shaping, and spread spectrum clocking (SSC) [1-4]. Among them, the spread spectrum clocking method is a simple and effective way to
reduce the EMI level [4]. The spread spectrum clock generator (SSCG) is a phase-locked loop (PLL)-based clock
generator that spreads the output clock signal energy concentrated in a narrow band
into a wider band spectrum. The modulation profile of an SSCG has a direct effect
on the spectrum spreading of the output frequency. Thus, an optimal modulation profile
is needed to reduce EMI reduction as much as possible [2]. In SSCG circuits, a triangle modulation profile is mostly adopted for its ease of
implementation. However, because most harmonic energies in the triangular modulation
profile are present on the edges of the spectral distribution, EMI reduction is not
optimal. Until now, a variety of design approaches for SSCG circuits with different
types of modulation profiles have been reported [3-15]. They mostly utilize a triangular modulation technique. It is known that a non-linear
Hershey-Kiss modulation profile gives more uniform spectrum distribution, and that
more EMI reduction can be achieved than with a triangular modulation profile. Implementation
of the nonlinear Hershey-Kiss modulation profile SSCG usually requires a somewhat
complicated design approach such as a look-up table approach [3] or a frequency-locked loop (FLL)-based SSCG with digital-to-analog converters (DAC)
and a delta-sigma modulator (DSM) [9]. Another SSCG with a nonlinear Hershey-Kiss profile was realized utilizing dual DSMs
[5]. In [5], the Hersey-Kiss modulation profile was obtained by connecting two up/down counters
and dual DSMs in series. The low EMI reduction of 15dB was measured in [5]. This is due to a large clock jitter induced by quantization noise from two DSMs
in series and an increased spectrum density near the center frequency during modulation
from the first order DSM.
In this paper, a revised approach is proposed to generate a Hershey-Kiss modulation
profile using two up/down counters and a single delta-sigma modulator. The first order
DSM for the slope control in [5] was removed, by which evenly distributed the spectral power density can be achieved.
This leads to a better EMI reduction. Adjustments of all design variables associated
with the proposed SSCG can be done digitally, thus the proposed SSCG has enhanced
portability for various applications and specifications. Additionally, a dual-tone
modulation profile was adopted to reduce further the EMI [6].
This paper is composed of five sections. In Section II, the design principle of the
proposed Hershey-Kiss modulation profile is explained. The setting of the design variables
and circuit design are described in detail in Section III. The measurement results
are given in Section IV and the final conclusions are presented in Section V.
III. THE PROPOSED SSCG CIRCUIT DESIGN
Fig. 2 shows a block diagram of the proposed SSCG, which is composed of a phase-locked loop
(PLL) and a digital modulation controller.
The PLL is a fractional-N PLL consisting of a phase-frequency detector (PFD), a charge-pump
(CP), a loop filter (LF), a voltage-controlled oscillator (VCO), and a multi-modulus
divider (MMD). The modulation controller consists of two Hershey-Kiss profile generators
(1 and 2) for dual-tone operation and a single delta-sigma modulator (DSM). The DSM
output is provided to the MMD to change the division ratio of the PLL periodically.
By varying the division ratio of the MMD, the output frequency is modulated and spread.
The dual-tone generator mixes outputs from two different profile generators. The main
difference from previous SSCG architectures is that the proposed SSCG is modulated
with Hershey-Kiss profile for achieving a better EMI reduction.
Fig. 2. Block diagram of the proposed SSCG.
1. Proposed Hershey-Kiss Profile Generator
Fig. 3 shows a block diagram of the proposed Hershey-Kiss profile generator. The slope profile
shown in Fig. 1(c) is to be generated. The profile generator has two UP/DN counters and accepts five
input signals to control the modulation parameters. The input signals include the
initial value of the slope (IS), number of discrete steps for slope control (M), change
of the slope per counting step (DS), peak counting value (K), and the spreading type
select (STS: down (‘01’), center (‘10’), up spreading (‘11’)). The outputs are a sign
(either positive or negative) and a slope value ($Out\_ slope$), which are provided
to the DSM. After initializing the output slope of the profile generator with IS,
UP/DN_counter_1 and UP/DN_counter_2 begin to operate. The peak counting value of K
sets the output range of the UP/DN_counter_1 and the counter output counts up and
down from ‘0’ to K and K to ‘0’, repetitively. The period of the counting value profile
relates to the targeted modulation frequency.
Fig. 4(a) shows the generation process of the proposed Hershey-Kiss frequency profile. The
profile generator produces a slope (Out_slope) having stairs-like discrete step patterns
(moving up or down) through UP/DN_counter_2 as the output of UP/DN_counter_1 changes
from ‘0’ to K, and vice versa, repetitively. UP/DN_counter_2 is for generating the
number of discrete steps of slope change over a quarter period of the modulation frequency.
If we use the more discrete steps for a linear frequency slope change over a given
period, the slope change on each discrete step becomes smaller. The number of discrete
steps of the slope can be adjusted by the external input M. UP/DN_counter_2 produces
a new pulse for every K/M pulse of the UP/DN_counter_1. By these newly produced pulses,
M different output slope values with step values (DS) are generated during a quarter
period of the modulation frequency. Fig. 4 shows the example of Out_slope when M is set as 5. The profile generator is not to
make more than M slope values when the counting value of UP/DN_counter_1 reaches K
or ‘0’. The sign value gives a positive or negative slope and is toggled whenever
the counter output value reaches K. When the value of the sign output is ‘0’, the
slope of the modulation frequency is positive, and the output frequency increases.
When the value of the sign output is ‘1’, the slope is negative, and the output frequency
decreases. If the sign output is ‘0’, the output frequency increases, and its absolute
slope is gradually decreased from the maximum to the minimum. Once the slope reaches
the minimum, then it is gradually increased from the minimum to the maximum. When
the counter output reaches K, sign switches to ‘1’ and the output frequency decreases.
Its absolute frequency slope is gradually decreased from the maximum to the minimum.
Once the slope reaches the minimum, then the slope increases from the minimum to the
maximum. The generated slope output makes the frequency variation smaller at the center
frequency and larger at the minimum and maximum frequencies over time. As a result,
the output frequency created is a Hershey-Kiss profile, as shown at the bottom of
Fig. 4(a). The continuous profile of the output frequency is shown with a black line. Based
on the discrete step value of Out_slope, the final profile is approximated as a piecewise
linear curve with a red line at the bottom (see Fig. 4(a)).
One period of the modulation profile consists of a positive slope swing and negative
slope swing. The shape of generated VCO control voltage is the same as the modulation
profile. Consequently, the period of UP/DN_counter_1 is the same as the half-period
of the sign output of the profile generator and the modulation frequency ($f_{m}$:
modulation frequency) is given by Eq. (1),
, where K is a peak counting value. If K is represented by a 10-bit counter and the
targeted modulation frequency is 33 kHz with a reference clock of 90 MHz, then K can
be set as 682. If the target modulation frequency is 30 kHz, K is set as 750.
As shown in Fig. 4(b), if the slope output (Out_slope) does not change (DS = 0), a typical triangular modulation
profile is generated because the frequency slope is constant. Proper IS and DS values
can be set on targeted spreading parameters of the SSCG. The maximum variation range
of the slope output value (D$_{M}$: | max. value – min. value | of Out_slope) during
a half-period of the modulation profile is given by Eq. (2).
The value (D$_{M}$) is provided to the input of the DSM. Because two Hershey-Kiss
profile generators were implemented, both D$_{M1}$ from the profile generator 1 and
D$_{M2}$ from the profile generator 2 are utilized for dual-tone mode operation. The
2-bit-size input signal (spread type select: STS) decides one of three spreading types,
which are down, center, or up spreading. When the external signal is ‘00’, ‘01’, ‘10’,
or ‘11’, the circuit operates in non-SSC, down-spread, center-spread, or up-spread
modes, respectively, as shown in Fig. 5.
Fig. 3. Proposed Hershey-Kiss profile generator.
Fig. 4. (a) Generation of the proposed Hershey-Kiss modulation profile (black line in Output frequency: continuous $Out\_ slope$, red line in Output frequency: discrete $Out\_ slope$); (b) Typical triangular modulation profile generation.
Fig. 5. (a) Non-SSC (STS = 00); (b) Down spreading (STS = 01); (c) Center spreading (STS = 10); (d) Up spreading (STS = 11).
2. Delta-sigma Modulator (DSM)
The delta-sigma modulator used in the proposed structure is the 1-1 structure 2$^{\mathrm{nd}}$
order multi-stage noise shaping (MASH) delta-sigma modulator. To optimize between
randomization and quantization noise effect of DSM, the second order DSM was adopted.
The DSM of 2$^{\mathrm{nd}}$ order MASH structure is stable. If a higher order DSM
were adopted, the phase jitter of the output clock could be increased.
Fig. 6 shows the MASH 1-1 DSM structure that takes outputs from the 15-bit Hershey-Kiss
profile generator output and produces 4-bit output. The 4-bit output of the DSM is
provided to the multi-modulus divider (MMD) in the PLL. The DSM generates three overflow
bits (C0, C1, and C2), which are mapped onto the 4-bit signal for MMD. As the overflow
bits in the modulator change, the average output of the DSM changes. Through the modulus
mapping circuit shown in Fig. 6, the values of C0, C1, and C2 are mapped onto 4 bits (R0, R1, R2, and R3) that are
provided to the multi-modulus divider. Transfer function of the DSM can be derived
as follows [16]. The overflow bit from a single accumulator with a feedback delay ($z^{-1})$ can
be expressed
Similarly, $C_{1}\left[z\right]=-e_{1}\left[z\right]+\left(1-z^{-1}\right)e_{2}\left[z\right]$
and $C_{2}\left[z\right]=-z^{-1}C_{1}\left[z\right]$. Therefore, $Y\left[z\right]$
can be derived as
Therefore, the signal transfer function (STF) and the noise transfer function (NTF)
given as Eqs. (5) and (6), respectively.
Table 1 shows an example of generation of the division ratio $N~ $between 54 and 57 by the
4-bit mapped outputs (R1, R2, R3, and R4) of DSM. Fig. 7 shows the MMD block diagram.
When the targeted spreading ratio (${\delta}$) is given, the value of ${\Delta}$N
can be derived. Once ${\Delta}$N is determined, the value of $D_{in}$ can be derived
with a given DSM bit-size. The relationship between the VCO output frequency and the
reference input frequency can be expressed as Eq. (7).
, where N is the targeted integer division ratio and ${\Delta}$N is the fractional
division ratio of the PLL. These are related to the targeted spread ratio of the SSCG.
The fractional division ratio ($\Delta N$) is the average value of the DSM output
and can be expressed by Eq. (8).
, where $D_{in}~ $is a DSM input. As shown in Eq. (6), if ${\Delta}$N is determined from a given spreading ratio (${\delta}$), the value
of $D_{in}$ can be obtained from Eq. (8). The procedure of setting design variables is explained in detail with an example
in section III-5.
Fig. 6. Second order MASH 1-1 delta-sigma modulator.
Fig. 7. Block diagram of the multi-modulus divider (MMD).}
Table 1. Modulus mapping circuit for N (55~57)
|
C0
|
C1
|
C2
|
R0
|
R1
|
R2
|
R3
|
N
|
|
0
|
0
|
0
|
1
|
1
|
1
|
0
|
55
|
|
0
|
0
|
1
|
0
|
1
|
1
|
0
|
54
|
|
0
|
1
|
0
|
0
|
0
|
0
|
1
|
56
|
|
0
|
1
|
1
|
1
|
1
|
1
|
0
|
55
|
|
1
|
0
|
0
|
0
|
0
|
0
|
1
|
56
|
|
1
|
0
|
1
|
1
|
1
|
1
|
0
|
55
|
|
1
|
1
|
0
|
1
|
0
|
0
|
1
|
57
|
|
1
|
1
|
1
|
0
|
0
|
0
|
1
|
56
|
3. Spread Ratio and Design Parameters
As shown in the previous section, the maximum variation of the output value ($D_{M}$)
during the half cycle of the modulation profile is given as Eq. (2). Now, the relationship between design parameters and the targeted spreading ratio
is described. The spreading ratio (${\delta}$) can be expressed as in Eq. (9), where N is the targeted integer division ratio and ${\Delta}$N is the fractional
division ratio of PLL, which is related to the spread ratio.
Therefore, substituting Eq. (8) into Eq. (9), the spreading ratio (${\delta}$) can be rewritten as Eq. (10).
, where $\left| D_{1}-D_{2}\right| $ is equal to the maximum variation of the slope
output value of the profile generator. Thus, $\left| D_{1}-D_{2}\right| $=$D_{\mathrm{M}}$.
For the targeted spread ratio, $D_{1}$ is set to an appropriate value. Therefore,
in the case of a single-tone modulator, the spread ratio can be written as in Eq.
(11).
In case a dual-tone profile generator is used, the spread ratio is expressed as Eq.
(12).
Because the slope output variation $\left(D_{M}\right)$ is determined by the initial
slope (IS), delta slope (DS), M value, and bit-size of DSM (as given in Eq. (2)), the spread ratio (${\delta}$) can also be adjusted with values of IS, DS, M, and
the DSM bit-size.
4. Dual-tone Modulation
As the spreading ratio increases, the frequency variation increases, and the EMI reduction
effect is increased. However, as the spread ratio increases, output clock jitter also
increases. As a result, this gives a design the complexity of having a clock and data
recovery circuit (CDR) in the receiver. If a single-tone modulation profile is used
with a fixed spreading ratio, the spectrum appears at the same interval as an integer
multiple of the modulation frequency. To reduce EMI, the frequency spacing at which
the spectrum appears needs to be reduced. A dual-tone modulation profile is a way
to reduce EMI further [6]. The dual-tone modulator is implemented by adding outputs from two profile generators
as shown in Fig. 2. The dual-tone modulators produce two different profiles of frequencies as shown
in Fig. 8. Due to the mixing of a low frequency component that changes the amplitude of the
output signal, the final spectral interval is not uniform. Thus, the spectral energy
is dispersed in a narrower frequency interval. As a result, a further EMI reduction
can be achieved. Ref. [6] produced a dual-tone profile with triangular shapes and achieves 18.8 dB EMI reduction.
The proposed SSCG approach adopted the dual-tone modulation approach from Ref. [6]. However, dual-tone Hershey-Kiss modulation technique instead of dual-tone triangular
modulation was applied and enables to achieve 28.7 dB EMI reduction. The combination
ratio (1-${\alpha}$, ${\alpha}$) of the two tones is made up by an appropriate combination
of peak values as shown in Fig. 8. The optimal ${\alpha}$ value was calculated to be about 0.86 [6]. Thus, when in dual-mode profile operation, the mixing ratio of 0.86 is used in the
proposed SSCG.
Fig. 8. Dual-tone modulation profiles[5].
5. Procedure for Setting the Design Parameters
In this section, the procedure for setting the SSCG design parameters is described.
First, assume that a 5~GHz output clock frequency is required with a 90 MHz input
reference clock. Then from Eq. (7), we can derive N = 55 and ${\Delta}$N = 0.5556, if an SSCG having a modulation frequency
of 33 kHz and a spread ratio of 0.5% is to be designed with a single-tone profile
generator. With a 15-bit DSM, $2^{SDM\_ bit\_ size}=2^{15}=32768$. Using Eqs. (7) and (8), the design variable of $D_{1}$, which is the maximum value of DSM input can be derived
as 18,205 with ${\Delta}$N = 0.5556. By substituting the modulation frequency of 33
kHz in Eq. (1), the peak counting value of UP/DN_counter1, $K_{1}$= 682. As M value is increased,
smoother modulation profile and a better EMI reduction is expected. Through simulations
with various M values, EMI reduction on different M values has been observed. The
simulations showed that little EMI reduction improvement has been achieved when M
is more than 5. Thus, M is set as 5 in this design. If we increase DS as 2, then IS
value is determined as 11. However, as the DS value goes higher, frequency variation
becomes smaller near the center frequency, so EMI reduction becomes smaller. Therefore,
for more even distribution of frequency variation, we chose DS = 1 in the proposed
SSCG design. Under various design constraints, IS and DS values can be optimized
for best EMI reduction through simulations. Setting $M_{1}$= 5, $DS_{1}$= 1, and $IS_{1}$=
9, the maximum variation of the profile generator, $D_{M1}$, is calculated as 9520.
From this, by plugging those values into Eq. (11), the targeted spread ratio of 0.5% is obtained as shown Eq. (13).
In case the dual-tone profile generator is applied, the second modulation frequency
of 30 kHz is added. The values of $K_{1}$ and $K_{2}$ are obtained using Eq. (1) at the given modulation frequency. From $f_{m2}=\frac{fref}{4\times K_{2}}=$ $30kHz$,
$K_{2}$ becomes 750 and $K_{1}$ is 682 at a modulation frequency of 33 kHz as before.
The other design variables are set as $IS_{1}$= 8, $DS_{1}$= 1, $M_{2}$= 3, $IS_{2}$=
2, and $DS_{2}$= 1 with the 15-bit DSM. The $IS_{1}$value fell from 9 (in a single-tone
profile generator case) to 8, so that the $\left| D_{M1}\right| +|D_{M2}|$ value was
adjusted at near 9600 for a 0.5% spread ratio. From the predetermined design variable
values, $D_{M1}$= 8160 and $D_{M2}$= 1500 can be obtained. Therefore, in the case
of the dual-tone profile generator, the final spread ratio $\delta _{dual-tone}=\frac{9660}{55\times
2^{15}+18205}\approx 0.5\mathrm{\% }\,,$which meets the targeted spread value.
IV. EXPERIMENTAL RESULTS
The simulation has been done with a 65 nm CMOS process on Hershey-Kiss modulation
profile generation. The design target is for generating the output clock with a center
frequency of 5 GHz using a 90 MHz input reference clock. The SSCG with modulation
frequencies of 33 kHz and 30 kHz and a target spread ratio of 0.5% was designed and
simulated. The design parameter settings applied to the SSCG circuit design are described
in section III-5. Fig. 9 presents a simulation result showing the single-tone Hershey-Kiss modulation profiles
of three spreading types. When the input STS values are ‘01’, ‘10’, and ‘11’, the
down-spread (Fig. 9(a)), center-spread (Fig. 9(b)), and up-spread (Fig. 9(c)) modes are shown, respectively. The simulation results of a single-tone modulation
profile (Fig. 10(a)) and dual-tone modulation profile (Fig. 10(b)) with a 0.5% spread ratio are also shown, respectively. In a single-tone modulation
case, the positive or negative peak value is constant over time. However, in dual-tone
modulation mode, the peak values are varied due to mixing two different modulation
frequencies. To show the difference between a single-tone modulation profile and a
dual-tone modulation profile, the x-axis scales are different in Fig. 10(a) and (b). The simulated phase noise is shown in Fig. 11. Our SSCG has a -92 dBc/Hz phase noise at 5 GHz with a 1 MHz offset.
To verify the proposed SSCG circuit, it has been fabricated using a 65 nm CMOS process.
Fig. 12 shows the layout of the proposed SSCG circuit and the chip image. The size of the
core block occupies an area of 231 ${\times}$ 191 ${\mu}$m, and the entire circuit
area including the loop filter is 735 ${\times}$ 530 ${\mu}$m. The following presents
measurement results of the proposed SSCG circuit with the design parameters in Section
III-5. Fig. 13 shows the measured frequency spectra at the output frequency of 5 GHz when operating
in single-tone profile mode. The resolution bandwidth (RBW) of the frequency analyzer
was set to 10 kHz.
The measurement results show that the peak power level of the SSC is reduced by 24.6
dB in down-spread mode, 24.4 dB in center-spread mode, and 23.9 dB in up-spread mode,
with a 0.5% spreading ratio. Fig. 14 shows the measured EMI reduction of the proposed Hershey-Kiss single-tone profile
and the dual-tone profile case. When operating in dual-tone mode, peak power reduction
of 28.7 dB is achieved. An additional EMI reduction of 4.1~dB was achieved in dual-tone
mode case compared to single-tone mode. Fig. 15 shows the measured output clock signal modulated with the dual-tone Hershey-Kiss
modulation. The rms jitter and peak-to-peak jitter of the output clock are 2.39 ps
and 16.4 ps, respectively.
Table 2 shows the circuit performance summary of the proposed SSCG design and comparison
with other recently published SSCG designs. The key performance is EMI reduction.
The performance comparison shows that the proposed SSCG circuit with dual-tone Hershey-Kiss
modulation has a better EMI reduction (28.7 dB) than in other SSCG work published
recently. In most of the SSCG works, EMI reduction measurements were done with a 10
kHz RBW. In terms of chip area, compared to Ref. [12] and Ref. [14] fabricated with the same 65 nm CMOS process, digital blocks for the dual modulation
profile generators. If loop filter of the proposed SSCG occupying about 85% of the
chip area and the placement of the other blocks were optimized, the chip area could
be reduced. The proposed SSCG chip consumes 8.5 mW for 5GHz clock generation with
dual-tone modulation operation and achieves 28.7 dB EMI reduction. Ref. [12] using the same process shows a 6.34 mW power consumption for 3.2 GHz clock and achieves
only 18.5 dB EMI reduction. And Ref. [14] generating the same 5GHz clock consumes 9 mW with 26~dB EMI reduction.
Fig. 9. Simulated single-tone modulation profile: (a) Down-spreading case; (b) Center-spreading case; (c) Up-spreading case (0.5% spreading ratio).
Fig. 10. Simulated modulation profile: (a) Single-tone modulation case; (b) Dual-tone modulation case (0.5% spreading ratio).
Fig. 11. The simulated phase noise at 5 GHz with frequency offsets of 10 kHz, 100 kHz, 1 MHz and 10 MHz.
Fig. 12. (a) Chip photograph; (b) Layout.
Fig. 13. Measured output spectra in three different spread types with single-tone modulation: (a) Non-SSC; (b) Down spreading; (c) Center spreading; (d) Up spreading.
Fig. 14. Measured output spectra: (a) single-tone case; (b) dual-tone case.}
Fig. 15. Measured clock jitter with dual-tone modulation.
Table 2. Performance summary and comparison with previous work