EE584 (Fall 2006) Introduction to VLSI CAD Project. Design of Ring Oscillator using NOR gates

Transcription

1 EE584 (Fall 2006) Introduction to VLSI CAD Project Design of Ring Oscillator using NOR gates By, Veerandra Alluri Vijai Raghunathan Archana Jagarlamudi Gokulnaraiyn Ramaswami Instructor: Dr. Joseph Elias

2 Table of Contents 1) Introduction ) The Ring Oscillator ) The Design ) Design of Ring Oscillator ) The Enable circuit ) ESD Protection Circuit ) Output Buffer ) Schematics ) Corner Testing Why corner testing? ) Layout ) Few points on the ring oscillator s layout drawing ) Few points on the output buffer s layout drawing ) Sweet/Bitter Experiences ) Conclusion References... 23

3 Table of Figures Figure 1 5 Stage Ring Oscillator... 4 Figure 2 35 Stage NOR Ring Oscillator... 6 Figure 3 Enable Circuit... 6 Figure 4 ESD Protection Circuit... 7 Figure 5 Output Buffer... 7 Figure 6 Output Buffer Circuit... 8 Figure 7 Overall Block Diagram... 9 Figure 8 Enable Schematic... 9 Figure 9 Ring Oscillator Schematic Figure 10 Output Buffer Schematic Figure 11 Slow slow Simulation plot Figure 12 Typi typi Simulation Plot Figure 13 Fast fast Simulation plot Figure 14 Slow Simulation Plot for 25 C Figure 15 Typi Simulation Plot for 25 C Figure 16 Fast Simulation Plot 25 C Figure 17 Slow Corner Simulation Surface Plot Figure 18 Typi Corner Simulation Surface Plot Figure 19 Fast Corner Simulation Surface Plot Figure 20 Overall Corner Simulation Surface Plot (Vdd = 1.8 V) Figure 21 Overall Corner Simulation Surface Plot (Temperature = 25 C) Figure 22 Ring Oscillator Layout Figure 23 Enable Circuit Layout Figure 24 Output Buffer Layout Figure 25 Overall Circuit inside bounding box... 22

4 1) Introduction The ring oscillator is a circuit used to calculate time delays in circuits and thereby help measuring the speed of a technology being developed. Since it is an oscillator it can also be used to clock other digital circuits. Designing a ring oscillator using NOR gates was the main goal of this project. Some other essential components like an enable circuit and an output buffer were also designed. The project dealt with a lot of learning, and understanding of VLSI concepts and practically implementing many ideas. 2) The Ring Oscillator The ring oscillator is a digital circuit consisting of many stages of inverters in series, where the number of stages is always odd. Figure 1 shows a simple five stage ring oscillator. When a digital 1 or a 0 is at the input of the first stage, it traverses through all the stages with alternate 1 s and 0 s at the output of every adjacent stage. As it has odd number of stages, the logic at the output is always the compliment of the input at the first stage and this output is fed back to the input thereby enabling oscillations. The time period for which the output is high or low is directly equal to the time delay (propagation delay) the signal experiences in traveling from the first stage to the last stage. Hence the period of the output waveform would be twice the time delay ideally. Figure 1 5 Stage Ring Oscillator 1 1 Page 339, CMOS Circuit Design, Layout and Simulation by R. Jacob Baker

5 3) The Design The primary focus would be to create a ring oscillator. The number of stages would depend on the delay one is expecting to get from the circuit. It should be noted that the circuit designed would be tested using an oscilloscope. A typical oscilloscope measures a frequency of about 100 MHz which corresponds to a time period of 10 ns. So under typical conditions it was decided to have a time delay of 15 ns, meaning a high time of 15ns and a low time of 15ns and hence a time period of about 30ns. This waveform can be easily measured in any standard oscilloscope. The motive was to design a ring oscillator with minimum number of stages and so the delay induced by each stage should be as high as possible. Hence the width of the transistors used in constructing the NOR gate should be minimum as larger width would result in more current flow reducing the delay in each stage. Also to have equal rise and fall times, the width of the PFET would have to be more than that of the width of the NFET to compensate for the different transconductance of electrons and holes. So the widths of all the PFETs were twice as that of the NFETs. After running a few trial simulations with NOR gates having minimum widths, it was observed that a 35 stage Ring oscillator using NOR gates would result in the desired time delay. Designing just a ring oscillator might not be enough as there are other things to be taken care of. The ring oscillator is a self starting oscillator and there must be a way to stop the oscillations whenever necessary. Hence an enable circuit is required to turn on and off the oscillator. Similarly the circuit needs protection from any kind of static discharge induced during testing. Hence an ESD (electro-static discharge) protection circuit is needed. Finally the ring oscillator must be able to drive some load. Since the circuit is designed for testing purposes, the load is taken as the oscilloscope. Keeping this load in mind, an output buffer circuit is also designed. These circuits are explained one after the other in the following sections. 3.1) Design of Ring Oscillator The objective is to create an inverter using a NOR gate. When the inputs of a NOR gate are shorted, there is only input and one output. Now the NOR gate behaves like an inverter. Thirty five such NOR gates were put in series to create the ring oscillator as shown in Figure 2. The ratios of the widths of PFET to the widths of NFET are shown above each NOR gate.

6 Figure 2 35 Stage NOR Ring Oscillator 3.2) The Enable circuit The enable circuit was taken from CMOS Circuit Design, Layout and Simulation by R. Jacob Baker. It is shown in Figure 3. It is a tri-state buffer where when the enable input is 1, the input goes to the output and when it is 0, the input is cut from the output and the output is a high impedance state. Figure 3 Enable Circuit 2 3.3) ESD Protection Circuit The ESD protection circuit is shown in Figure 4. Diodes D1 and D2 are used to clamp the input signal to within 0.7 V of VDD or ground. The resistor R is used to further minimize any input current in the event D3 breaks down or conducts. The resistor is usually realized by a long n+ diffusion region that might surround the contact pad. 3 2 Page 372, CMOS Design, Layout and Simulation by R. Jacob Baker 3 Page , Digital Integrated Circuit Design by Ken Martin

7 4 Figure 4 ESD Protection Circuit 3.4) Output Buffer As mentioned in the design, the circuit must be able to drive a load. In this case the load is assumed to be the oscilloscope. After going through a few datasheets of some oscilloscopes, it was found that the oscilloscope usually has a load comprising of a capacitor 20pF and a resistance of 1 mega ohm. So the ring oscillator must be able to drive it and hence we need an output buffer. The design of the output buffer is got from CMOS Design, Layout and Simulation by R. Jacob Baker and is shown in Figure 5. The explanation given is Consider the inverter string (a buffer) driving a load capacitance, labeled C load and shown in Figure 5. Moving towards the load in a cascade of the N inverters, each inverter larger than the previous by a factor A (that is the width of each MOSFET is multiplied by A), a minimum delay can be obtained as long as A and N are picked correctly. Each inverter s input capacitance is larger than the previous inverter s input capacitance by a factor of A. 5 The equations for designing the circuit are as follows: N=ln(C load /C in1 ) and A=(C load /C in1 ) (1/N) 6 C in1 is the input capacitance of the first stage inverter. Figure 5 Output Buffer 7 4 Page 274, Digital Integrated Circuit Design by Ken Martin 5 Page 344, CMOS Design, Layout and Simulation by R. Jacob Baker 6 Page , CMOS Design, Layout and Simulation by R. Jacob Baker 7 Page 344, CMOS Design, Layout and Simulation by R. Jacob Baker

8 The first inverter is assumed to have p-width/n-width of 2.10/1.05. The C in1 value is obtained by finding the C oxp and C oxn for the PFET and NFET. C in1 =3/2*(C oxp +C oxn ) C oxp = C oxp 1 *W*L*(scale) 2 8 C oxn = C oxn 1 *W*L*(scale) 2 9 Where, Cox1 is got from Table 6.4, CMOS Circuit Design, Layout and Simulation by R. Jacob Baker W Width of the transistor L Length of the transistor Scale 50 nm (all parameters multiplied by scale to get actual values) Solving all the above equations the value is N is obtained as 10 and value of A as e. If N is 10 then the widths of the PFET and NFET for the 10 th stage would be very large and difficult to implement in layout. So the buffer is stopped with just 9 stages as this would be good enough, if not the best to drive our desired load. There is a stage before the first stage of the actual buffer circuit. This inverter is added to make the total number of stages even so that the output of the ring oscillator appears exactly at the load. If there are 9 stages then the output at the load would be the inverted version of the output of the ring oscillator. This design is shown in Figure 6. The widths of the PFET/NFET are given above each stage. Figure 6 Output Buffer Circuit 8 Figure 6.5, CMOS Circuit Design, Layout and Simulation by R. Jacob Baker 9 Figure 6.5, CMOS Circuit Design, Layout and Simulation by R. Jacob Baker

9 The overall design is shown in Figure 7. Figure 7 Overall Block Diagram 4) Schematics Once the design has been decided, the next thing to do would be to draw the schematic. This was successfully done using the Cadence software. Some of the snapshots from Cadence are included below. Figure 8 shows the schematic of the enable circuit, Figure 9 shows the ring oscillator structure without any feedback, Figure 10 shows the output buffer with the series of inverters, and Figure 10 shows the ESD protection circuit. Figure 8 Enable Schematic

10 Figure 9 Ring Oscillator Schematic Figure 10 Output Buffer Schematic

11 5) Corner Testing Once the schematics of the various circuits are drawn, the most important thing is to make sure that the design actually works. Often, system being designed might not work properly and if the layout for such a design is made then it would be a total waste of time and money. So, the only way to avoid that is to run simulations and check if the desired results appear. So, simulations are carried out in various temperatures to see how the output varies with temperature. Often in digital circuit the input voltage fluctuates a little mainly due to noise in the power lines. So the design must be tested under these conditions also. Usually the voltage variation is within 10% of its actual value. So the voltage was varied in this 10% range and the corresponding changes in the output were noted. While conducting simulations, the design has to be tested in various corners. Extreme conditions in which the design is put to test are usually called corners. In this case the extreme conditions are fast and slow, and these are achieved by varying certain internal parameters like oxide thickness. The fast corner refers to the case where the current flow and switching speed are very fast, while the slow corner refers to slower current flow and switching speed. There is also the presence of a typi condition in which all the internal parameters are set to typical values. Why corner testing? Often during the process of manufacturing the desired circuit, certain internal parameters might change a little. There might be a slight deviation in these parameters from the typical values. For instance, during the process of manufacturing the oxide thickness might change by couple of nanometers. This change might induce a lot of variations in the output generated by the design. So to make sure that the design is stable even at the corners, corner simulations are conducted. These simulations also give the designer an idea of whether his/her design would work in all the conditions and if it does not work, why it does not work and under what conditions it does not work. So it is an useful tool to analyze the design. Some of the simulation results are shown below. Figure11 shows the variation of time period (the time delay) with respect to temperature for various voltages in the slow corner. From the plot, it is evident that for a particular voltage, as temperature increases the time delay decreases. This phenomenon is present for even the fast corner and the typical conditions as shown in Figures 12 and 13. But it has to be observed that for a given temperature and voltage, as the simulation environment varies from slow to fast, the time delay decreases. This is very much in line with the expected results.

12 Slow_Slow Simulation Time Period (n Temperature (C) 1.8V 1.82V 1.86V 1.9V 1.94V 1.98V 1.74V 1.7V 1.66V 1.62V Figure 11 Slow slow Simulation plot Typi_Typi Simulation Time Period (n Temperature (C) 1.62V 1.7V 1.8V 1.9V 1.98V Figure 12 Typi typi Simulation Plot

13 Time Period (n fast_fast Simulation Temperature (C) 1.62V 1.66V 1.7V 1.74V 1.8V 1.82V 1.86V 1.9V 1.94V 1.98V Figure 13 Fast fast Simulation plot Though it was not very necessary to plot the relation between voltage and time period for a constant temperature, it was plotted to give a better understanding of certain concepts. Figure 14 shows the plot between voltage and time period for a constant temperature of 25C in the slow corner, Figure 15 shows the same plot under typical conditions and Figure 16 shows the same plot in the fast corner. It can be observed that for a constant temperature, as the voltage increases the time delay decreases. This is in accordance with the fact that more voltage could drive more current through a device. As current through the device increases the time delay induced decreases because the input capacitance of the device is charged and discharged faster. The reduction in time delay with change in the corner from slow to fast can also be observed (for a constant temperature and voltage).

14 Slow 25C 250 Time Period (n C Voltage (V) Figure 14 Slow Simulation Plot for 25 C Typi 25C Time Period (n Voltage (V) 25C Figure 15 Typi Simulation Plot for 25 C

15 Fast 25C 25 Time Period (n C Voltage (V) Figure 16 Fast Simulation Plot 25 C Figures 17, 18 and 19 show the combined result of voltage, temperature and time period in the same surface plot for slow corner, typical condition and fast corner 'slow' Corner Simulations voltage(v) Temperature(c) Time period(ns) Figure 17 Slow Corner Simulation Surface Plot

16 'Typi' Corner Simulations Temperatu re(c) Period(ns) Voltage(V) Figure 18 Typi Corner Simulation Surface Plot 'fast' Corner Simulation voltage(v) time period(ns) temperature(c) Figure 19 Fast Corner Simulation Surface Plot

17 Figures 20 and 21 show the overall result of all the simulations that were conducted. Figure 20 shows the variation of time period with the variation in temperature for each of the three corners, at a constant voltage. Corner Simulation (VDD = 1.8V) Temperature (c) 85 slow_slow typi_typi fast_fast 150 Time Period (ns) Figure 20 Overall Corner Simulation Surface Plot (Vdd = 1.8 V) Figure 21 shows the variation of time period with variation in voltage for all the corners at a constant temperature.

18 Corner Simulation(Temp = 25 c) Time period (ns) slow_slow typi_typi fast_fast Voltage (V) Figure 21 Overall Corner Simulation Surface Plot (Temperature = 25 C) The following two tables show the data that were used to plot the surface plots in Figure 20 and 21. Voltage = 1.8V Temperature Time Period (ns) slow_slow Typi_typi Fast_fast Temperatre = 25C Voltage (V) Time Period (ns) slow_slow Typi_typi Fast_fast

19 After going through the simulations it was observed that the design worked well in all the simulation corners, as per the predicted results. But it should be noted that in the fast corner, when the temperature range is high, then the design s output cannot be viewed in an oscilloscope as the time period reduces below 10 ns, especially for voltages from 1.8V onwards. Similarly the time delay increases a lot in the slow corner especially when the temperatures are low and voltages are also less. One would not prefer operating the device in that range. 6) Layout After the simulations have been conducted successfully, it is known that the designed circuit works and it can be implemented. So the next task would be to draw the layout and make sure the layout passes all the software design rules. Figure 17 is the snapshot of the layout of the ring oscillator, Figure 18 is the snapshot of the layout of the enable circuit, Figure 19 is the snapshot of the layout of the output buffer and Figure 20 shows the layout of the entire circuit inside the bounding box. 6.1) Few points on the ring oscillator s layout drawing While drawing the basic NOR layout used in the ring oscillator (RO), it was made sure that input and output lines were extended such that they don t go beyond the N well. This is done to make sure that metal is not used again to join them. The ring oscillator s layout was done making use of rows and columns available in the instance tool box. Metal crossing was eliminated by routing metal1 to licon and crossing it over another metal1.

20 Figure 22 Ring Oscillator Layout Figure 23 Enable Circuit Layout

21 Figure 24 Output Buffer Layout 6.2) Few points on the output buffer s layout drawing A primary inverter cell was created using a ptran and ntran with the corresponding widths and lengths. The area of the bounding box was considered (2700*125 micrometers) and the values of the cascaded stage s areas were decided so as to fit into this. The maximum resolution in Cadence (0.025) was used. It was found that every stage except for the last stage was very small compared to the area between any two e-pads. The last stage had a large length (not channel length) such that it could only fit between 2 e-pads if placed perpendicular to the long sides of the bounding box. Since there is no restriction on placing the last stage in the bounding box (considering facts like connections to VDD and GND and there is only one output connection), it was placed between the two most extreme e-pads on one side of the bounding box. As the rest of the stages are small they could be moved closer to the longer sides of the bounding box along which the power buses were placed. Since there was only one connection between the last stage and the remaining circuit, a whole e-pad was used as a junction rather than routing, which causes parasitic and waste metal. For a CMOS inverter, it is simple to bring the input and the output terminals on either side of the inverter. This fact was used to simplify interconnecting stages. Strapping was done till the met1 layer for the following reasons. Met1 is the lowest layer that can be used to interconnect devices through low resistance. Also LI has severe DRC restrictions regarding the proximity to diff areas.

22 Length of met1 contacts on transistors made it easy to connect a power bar perpendicular to them to it, and in most cases no metal patches were created to connect the power bar to met1 contacts from transistors. There is no need to rise the routing to met2 unless there is a crossing of two met1 routes. In such cases it was used. Figure 25 Overall Circuit inside bounding box 7) Sweet/Bitter Experiences While doing the project, there were a lot of good experiences and some bad ones too. The good experiences are as follows. The prescribed book CMOS Circuit Design, Layout and Simulation by R. Jacob Baker was very useful while designing the entire circuit. In Cadence software, the tech library had pre-defined schematic and symbols for the inverter, two input NAND, and NOR. These symbols were directly used in the design and it saved a lot of time and work. While running simulations, there was this option called parametric testing which eased the simulation process as the simulation could be made to run with a variation in many parameters in a single run.

23 The m-factor option in Cadence software for the transistor folding was very helpful especially for the layout as the dimensions of the transistor could be suitably varied in order to make it fit inside the bounding box. The bad experiences are as follows. Even though the parametric analysis was a useful tool, it was extremely slow in producing the final result and sometimes it failed to produce the result, especially when the temperature was negative. The drawing of the layout was very tough and there were a lot of errors encountered while trying to pass through the design rules. Some of these errors were due to minimum spacing between nwells, between nwell and diff, between poly and poly, met1 and met1. Drawing the layout required a lot of patience and concentration, as the design rules had to be passed and several other factors such as minimizing the layout area, having less resistance while routing and so on had to be considered. 8) Conclusion The various steps involved in the project are as follows. First, a suitable design with all the circuits was chosen and it was decided to implement the design. Then, the schematic of the various components in the design were drawn and then connected properly. Next, to make sure that the design was proper and would work correctly, a lot of simulations were performed based on the schematic. Once the simulations yielded good acceptable results, the layout for the design was drawn. The layout was made to pass all the required design rules and LVS successfully. The project provided an opportunity to learn how the various parameters and settings actually influenced the design. It also provided information on how different kinds of inputs yielded different outputs. A lot of practical problems were analyzed while performing the various tests and simulations. The project was a good learning experience. References CMOS Circuit Design, Layout and Simulation by R. Jacob Baker. Digital Integrated Circuit Design by Ken Martin.