ELEC4602
Microelectronics
Design and Technology
Lab 4: Combinational logic design
Objective
The objective of this laboratory session is to work with a somewhat larger combinational digital circuit, namely a full-adder, and to design and characterise this adder. For the hand-calculations, you should use the parameters for the NSCU FreePDK 45nm CMOS process published at the course web-site. The hand-calculations should not be carried out in the laboratory session!
Full-adder design
The full-adder symbol and logic function is shown in figure 1. An example (from Baker) of a static CMOS implementation of the full-adder is shown in figure 2; you may use this schematic or use your own. The schematic in figure 2 directly implements the logic function in figure 1: the node CO in the first schematic, for instance, is either pulled to VDD through a series of PMOS transistors, or pulled to ground via a series of NMOS transistors. Looking at the NMOS pull- down network, we can see that CO is pulled to ground if A and B are both high (causing both bottom-left NMOS transistors to conduct) or CI is high and either A or B is high; looking at the PMOS pull-up network, we can see that CO is pulled to VDD if A or B islow and CI is low or both A and B are low; thus the logic function is realised. You should create a cell that contains the schematic of your chosen full-adder, and use test-benches as appropriate for for the simulations that you are going to carry out.
Figure 1: Full-adder symbol (a), known parameters and functions (b)
Transistor types: In this lab, regular thin-oxide transistors (N VTG and P VTG) are used for all transistors.
Figure 2: Static CMOS example implementation of full-adder
Transistor dimensions: As noted in the figure, all transistor dimensions should have a lengths of L = 0.05 µm; you need to choose the widths for all the transistors: if there is no particular load requirement to a logic gate, the transistors should be chosen as small as possible; while the smallest allowed transistor width is 90 nm, it is most convenient from a layout point of view to choose transistor widths as an integer multiplum of theminmum contact pitch (i.e. WunitN = 65 nm + 75 nm = 140 nm), so choose this as the unit width for NMOS transistors. The PMOS transistor unit width is usually chosen one to three times the NMOS transistor one. The unit widths are the dimensions used in a unit size (or “normal”) inverter. For more complex gates, like the ones shown in figure 2, transistors are normally increased in width when they appear in series, in order for the delay through the gate to be similar to that of the unit inverter. The normal sizing approach is this: if a transistorsit in a branch with N series transistors between the output and the relevant power supply (ground for NMOS and VDD for PMOS), the transistor width is chosen as NWunit.
Test set-up: The power supply used in this session is VDD = 0.9 V; this also means that digital values are either 0 V (0) or 0.9 V (1). In your test-bench, use load capacitances at both full-adder outputs (S and CO) of CL = 8 fF. Rise and fall times of about 50 ps and data rates about 109 s-1 are probably appropriate.
Verifying logic function
To verify the logic function of your full-adder, apply input voltage waveforms that cycle through all eight input combinations (for instance using vpulse voltage sources with different periods – e.g. 3 ns, 6 ns and 12 ns – at the three inputs and running a transient simulation; see figure 3) and check the outputs to see that the logic function implemented is the desired one.
Finding propagation delays
If the three inputs waveforms are arranged such that they never change state at the same time (as suggested in figure 3), you should be able to find relevant propagation delays in your simulation.
Figure 3: Suggested input waveforms for full-adder
The Delay time property in the vpulse voltage source is useful for this: arrange your inputs to be delayed, say, 0 ns, 0.5 ns and 1.0 ns. Now, making sure that a change in CI causes a change in CO, find the H→ L and L→ H CI to CO propagation delays.
Finding power dissipation
The power dissipation in static CMOS logic is almost exclusively dynamic; i.e. caused by charg- ing parasitic capacitances, and to some extend crow-bar currents. The power dissipation is equal to the power being delivered by VDD (be sure to have only one VDD voltage source in your schematic; the one living in your test bench):
where the integral finds the average current delivered by VDD . Now, find the power dissipa-
tion in your full-adder: in the transient simulation above, plot the current in the power supply
VDD (probably called /V1/MINUS or something similar). To find the time integral of this cur-
rent, in the graph window, select Tools-Calculator which will open up a graph Calculator window. In this, clear the “buffer” entry field by clicking on the Clear buffer button (or just
use backspace); then, in the graph window click on the supply current name (/V1/MINUS); that
should enter this waveform. in the calculator (probably looking something like IT("/V1/MINUS/")). Now click on iinteg (indefinite integral) in the Special Functions fieldof the Calculator (or just
edit the buffer to read iinteg(IT("/V1/MINUS"))), and click on the Evaluate the buffer button (you can choose to have the trace appearing in a New Window if you so choose); this should plot the time integral of your supply current in the graph window. In the graph window, choose Marker-Create Marker . . . and click OK in the marker creation window. Repeat this step to create another marker. Now select both markers and choose Marker-Create Delta Marker. Move one marker to the start of your time integral signal and the other to the end – the delta marker should now read the average slope of your time integral, and hence the average supply current from which you can calculate the power dissipation.
Corner simulations
The simulations done so far has been carried out using a typical set of process parameters. In reality, there is a lot of variation between different wafers fabricated. For digital circuits, it is usually regarded as sufficient to simulate in two worst-case corners (in addition to the typical parameter values), namely the slow corner (low gain, high capacitance) and the fast corner (high gain, low capacitance). To simulate your circuit in the slow corner, in the Virtuoso ADE Explorer window maestro menu choose Setup-Model Libraries . . ., in the Model Library Setup win- dow, replace the model library that you are using to the on ending in . . ./hspice_ss.include (as opposed to . . ./hspice_stat.include), and click OK. Simulations done now are carried out in the slow corner for the process. Use the same approach carry out simulations in the fast corner (replace model library with the one ending in . . ./hspice_ff.include). Find the H→ L and L→ H CI to CO propagation delays in the slow and the fast corner.
Driving a large load
When driving a large capacitive load from a unit sized gate, it is necessary to inserts buffer in- verters with increasing drive strength (i.e. transistor widths): The NMOS transistor in an inverter of drive strength S has a width of SWunitN while the PMOS transistor has a width of SWunitP . The usual approach is shown in figure 4 where two inverters are inserted after the unit inverter such that the load capacitance CX can be driven to the digital value X in reasonable time. The scale factor k (drive strength ratio between consecutive inverters) is normally in the range 3 to 6.
Change the load capacitance CLS at the S output to 150 fF, and insert suitable buffer inverters, such that this load can be driven at the data rate you have been using in the other simulations; verify this by a transient simulation.
Figure 4: Unit inverter driving large load; inverters are labelled with drive strengths
Report
A short report in .pdf format on the laboratory exercise must be prepared and uploaded on the course Moodle site no later than the due date. This need to include:
• The schematic of your test bench.
• The schematic of full-adder (explain choice of transistor widths).
• Proof that S is always pulled to ground or VVDD (but never both).
• Transient simulations showing logic function.
• Transient simulations showing the CI to CO propagation delays in the fast and the slow corner (compare this with simple hand calculations using the typical parameters).
• Simulations used for power dissipation calculation, and a calculation of the power dissipa- tion.
• The schematic of the buffer inverters for driving a large load on S.
• Transient simulations showing proper operation when driving a large load on S.