This preview shows page 1. Sign up to view the full content.
Unformatted text preview: Massachusetts Institute of Technology
Department of Electrical Engineering and Computer Science
High Speed Communication Circuits and Systems
Project #1: Design of a High Speed Divide-by-32/33 Prescaler
Passed Out: March 21, 2003 Due: April 11, 2003
Copyright c 2003 by Michael H. Perrott 1 Objective This project will introduce you to the art of designing high speed frequency dividers. Your
goal will be to design a divide-by-32/33 prescaler, as shown in Figure 1, that supports a
minimum input frequency of 5 GHz with maximum power dissipation of 10 mW. Calculation
of power dissipation must include all blocks in Figure 1 (including bias sources) except for
input voltage sources Vin and Vin . You should strive to exceed one of these speciﬁcations
• You achieve an input frequency higher than 5 GHz with power dissipation not exceeding
10 mW, OR
• You achieve power dissipation less than 10 mW with an input frequency equal to or
above 5 GHz.
In other words, you should explore either a high speed or low power approach under the
rise/fall time = 40 ps
50 Ω Vin Vin 1.8 V
5 GHz 32/33 IN OUT
Reg CON DQ Figure 1: High Speed Divide-by-32/33 Circuit. 1 OUT
Cload=100 fF 2 Report Your report should consist of the sections described below. Note that one report is turned in
for each project team (consisting of exactly 2 people), and must be no longer than 15 pages
including all plots and ﬁgures.
• A brief introduction of the project and a statement of your overall goal (i.e., a high
speed or low power divider design)
• A brief summary of the performance you achieved, which includes a table of the values you obtained for the following speciﬁcations (assume a temperature of 25 degrees
Celsius except where otherwise stated):
1. Maximum input frequency of overall divider
– At 25 degrees Celsius
– At 0 degrees Celsius
– At 80 degrees Celsisus
2. Maximum input frequency achieved by each major section of the divider (i.e., the
front end divider circuit, the diﬀerential-to-full swing circuit (if used), and the
back end divider circuits)
3. Total power dissipation (including bias sources)
4. A breakdown of power dissipation in the divider (i.e., list the power consumed for
each major block in the divider)
5. Overall speed/power metric for the entire divider (i.e., Maximum speed (GHz)/Power
• A brief summary of simulated results that includes Hspice plots of the divider input
and each divider stage output at the highest working frequency (with temperature set
to 25 degrees Celsius). The plots must clearly show both the divide-by-32 and divideby-33 operations as part of the same simulation run (i.e., you must switch the value of
division during the simulation).
• A brief summary of the merits of your approach, which includes the following two
– The beneﬁts and shortcomings of your architectural approach to the divider (i.e,
why is it good for attaining high speed, or low power operation?) Also, explain
the bottleneck in your architecture that prevents achievement of even higher input
– A brief description of the function and performance tradeoﬀs for each circuit
block. (Example: ﬁrst stage performs such and such operation, and was designed
for high speed by doing so and so. Power dissipation was minimized by doing so
2 • A brief, qualitative summary of the impact of temperature variations on the divider
performance. Speciﬁcally, is the performance better or worse at low and high temperatures, and why? (Note that the quantitative impact of temperature variations on
divider speed is speciﬁed in the speciﬁcation table described above.) In practice, one
would also need to account for variations in transistor performance (i.e., using slow
and fast models) and for resistor variations (typically ± 20 %), but we will ignore these
issues in this project for the sake of simplicity.
• Conclusion 3 Constraints The following constraints must be met when stating results of your divider performance
• The model ﬁle /mit/6.976/Models/0.18u/mos018.mod must be used when simulating
NMOS and PMOS devices
• Minimum length of all devices is 0.18 µm, and minimum width should be assumed
to be 0.5 µm. (Note that the actual process would allow smaller widths, but these
wouldn’t be practical if wiring capacitance was included (i.e., the overall capacitance
would then be dominated by the wires). We are not considering wire capacitance in
this project, but we do want to be realistic in our consideration of its impact on sizing)
• Source and drain junction capacitances must be included for all transistors by specifying the perimeter and areas of each source/drain junction. You should use hspc for
this as described on page 6 of the hspc manual (at
http://www-mtl.mit.edu/research/perrottgroup/tools.html), where hdin = 0.4 µm, hdout = 0.5µ m, and a four sided perimeter should not be assumed. You are free to make
use of ﬁngers (as speciﬁed by the m and geo parameters) to reduce the impact of
• Capacitance associated with poly resistors (in SCL logic) must be included in the
simulation. The overall capacitance of a given resistor, Cp , is calculated based on
its area and perimeter values, which are set by its resistance value and the required
average current it must handle. Given the area and perimeter of the resistor, the overall
capacitance is then calculated using the area and perimeter capacitances of poly over
ﬁeld (i.e., substrate).
To perform the above calculations, the relevant parameters for unsilicided P-doped
poly (which will be assumed to be used for resistors in this project) are
– Area capacitance from poly to ﬁeld: 100 aF/µm2
– Fringe capacitance from poly to ﬁeld: 50 aF/µm2
– Sheet resistance of poly: 150 Ohms/square
– Electromigration rule for poly: ≤ 1 mA/µm (applies to average current)
3 – Minimum poly width/length for resistors is 0.5 µm (to avoid dogbone layout of
resistors when contacts are included)
Finally, a distributed model should be assumed for the impact of the total resistor
capacitance, Cp , as shown in Figure 2.
R/2 R/2 Cp/4 Cp/2 Cp/4 Figure 2: Modeling of distributed capacitance for resistor load.
• The maximum supply voltage is 1.8 Volts.
• The input to the divider is assumed to consist of two signal sources with 50 Ω source
resistances, Vin and Vin , which are 180 degrees out of phase from one another and have
the following properties
Vhigh = 1.8 V, Vlow = 0.9 V, risetime = 40 ps, falltime = 40 ps
Note that 5 GHz versions of these signals can be achieved using the hspc program with
the following lines
% timing 0n 40p 100p 0.9 1.8
% input vin [set 1 0 R]
% input vinb [set 0 1 R] • The output to the divider must drive a load capacitance of 100 fF and a register that
clocks in the CON signal ( on either its rising of falling edge — your choice) as shown
in Figure 1.
• You are allowed one ideal current source of any value that you choose (you must include
its power dissipation when specifying the overall power dissipation of the divider). The
source may be hooked between any nodes of your choice so long as one of the nodes is
Vdd or ground.
• Inductors can not be used in this project (in practice, they yield better performance
at the cost of signiﬁcant die area). 4 4 Grading The project will be graded as follows
• Understanding (10 %) — does your description of the architecture and circuits reveal
that you understand the basic principles of designing a divide-by-32/33 circuit? Why
did you choose the architecture, the type of logic style that was used in diﬀerent parts of
the divider, and did you clearly state the bottlenecks to achieving better performance?
• Originality (20 %) — is your divider implemented in a unique way that oﬀers some
• Performance (30 %) — how fast does your divider operate (if speed was your goal)?
How much power does it consume (if low power was your goal)?
• Practicality (10 %) — Are there any issues that would prevent your design from being
• Description (20 %) — are the fundamental concepts of your design clearly explained?
Are the advantages of using your approach clearly spelled out? Is your writeup concise
and to the point?
• Simulation (10 %) — did you adequately verify the performance of your design? 5 Suggested Design Procedure Determine the achievable speed of a divide-by-2 conﬁguration implemented with several
diﬀerent logic styles. Try to understand why some logic styles perform better than others.
(A good, brief description comparing a few logic styles you tried and explaining the reasons
for their performance diﬀerences will boost the grade you receive in the “Understanding”
category.) Repeat this process for a divide-by-2/3 circuit.
Once you decide on logic styles for diﬀerent parts of the divider, cascade the resulting
divide-by-2/3 with divide-by-2 sections and then design the control qualifying logic to achieve
proper timing of the control signal to the divide-by-2/3 stage. Verify that the control qualifying logic works properly by varying the divide value in a SPICE simulation and verifying
that pulse swallowing occurs at the proper time.
Note that phase shifting techniques will achieve higher speed than gated approaches to
the divide-by-2/3 implementation, but are much more diﬃcult to design. I would suggest
starting with a gated design, and then tackle phase shifting if you have the interest and time. 5 ...
View Full Document
- Spring '03
- Electrical Engineering