Neuromorphic Engineering II Lab 2, Spring 2013 Lab 2 1 February 26, 2013 SPICE The objectives of this lab are to: • Learn how to use SPICE for transient (time-stepping) circuit simulations. Unlike AC simulations, which are linearized around a DC operating point, transient simulations use the full nonlinear device model. As a result they are much slower but can reveal large signal and transient (e.g. startup) behavior. They are also the only device-level simulations possible for unstable or multistable circuits like clock generators, neuron circuits, or flip-flops. • Simulate an integrate and fire neuron circuit. This week, we will use the software tool S-Edit to enter schematics and T-Spice to run transient circuit simulations. 2.1 Reading Read the lecture notes and revert to the online TSpice documentation (PDF files) when looking for details. 2.2 Prelab 1. The prelab should be completed before any computer is touched. 2. This exercise asks you to write a spice netlist by hand. It is a tedious excercise, but very useful. a. Use the circuit diagram of Fig. 2.1 and write the circuit netlist using your favourite text editor. b. Use the labels nmos and pmos for the transistor models c. Use a value of 500fF for the Capacitor. d. As we will use the scale option, in the simulation, use dimensionless values for L and W (e.g. W=12, L=6). e. Save the file as myaxonhillock.sp Neuromorphic Engineering II Lab 2, Spring 2013 2 Figure 2.1: Axon Hillock circuit. f. Using the neuron circuit in Fig. 2.1, reason what happens to the membrane potential, Vmem, when a DC input current is injected in the node Vmem itself, and charges up the membrane. Also explain how the spike output Vout of the circuit behaves. 2.3 Simulation using Tspice We will use the Tanner Tools to simulate the neuron circuit in Fig. 2.2. This is the first VLSI integrate-and-fire circuit that was proposed by Carver Mead and colleagues in the late ’eighties. 2.4 Experiments The SPICE simulation will be run directly from within S-Edit: S-Edit can generate spice files with both schematics and commands, in spice-format, and pass them to T-Spice. You will learn to generate these files automatically, but also edit them by hand. Experiment 2: Simulating the axon-hillock circuit You will be running simulations from a file in which most of the work has already been done for you. But you are requested to take notes of the simulations and hand-in a report in which you print all the anwsers to the questions below and include the plots requested. In the report, include the code of the “myaxonhillock.sp” file that you made in the prelab. 1. Open the file exercice3 provided by the TA. Neuromorphic Engineering II Lab 2, Spring 2013 3 Figure 2.2: Axon Hillock circuit. 2. Open the cell “axonhillock” and study the circuit. 3. Study all of the options and settings in the Spice setup diag box (from the menu Setup→Spice Simulation). What do the options in the “Additional spice commands” mean? (hint: you can study the t-spice manual from the t-spice help menu.) 4. Run a Spice simulation, and study the output. Are the output waveforms like you expect them to be from the pre-lab? Explain why (or why not). 5. Re-run the simulation with the “power-up” option in the transient analysis. Ho do the simulations differ? What does this option do? (hint: see the t-spice manual.) 6. Change the value of the feedback capacitor and experiment with values ranging from 10fF to 500fF. Run the simulations for at least 3 different value settings. What changes in the simulations? Can you eplain why? 7. Change the input current and re-run the simulation until you get 10 spikes in total. Produce a plot the waveforms on separate windows, with a zoomed version of just 2 spikes (and include it as a figure in the report). 8. Now switch to T-Spice, and open the file “axonhillock.sp” from within the main TSpice window. Study it, and see if the main structure differs from the spice-file you wrote by hand. From the t-spice manual find out how to measure the average current flowing through the n-FET of the first inverter connected to Vmem, in the transient analysis. Re-run the simulation, find the measurement and note it in the report. The result of the measurement will be toward the end of the “axonhillock.out” file. 9. Find out how to measure the average power dissipated through the Vdd voltage source, re-run the simulation, and print in the report the result obtained. What percentage of the total power budget is dissipated by the first inverter? (hint: use the current measurement made above to compute it). Neuromorphic Engineering II Lab 2, Spring 2013 4 Figure 2.3: Axon Hillock circuit. Experiment 3: Low-power silicon neuron Here you will run simulations on a new silicon-neuron circuit, and compare its power dissipation with the one of the standard axon-hillock circuit. 1. Copy the “axonhillock” into a new cell (e.g. soma lowpower). 2. Modify the new cell, and create a circuit exactly as shown in Fig. 2.3 (including the values of the capacitors and biases). 3. Run the simulation and verify that the circuit works as a silicon neuron. 4. Find the value of the input current that produces the same number of spikes used in the axonhillock simulations, where you computed the average power dissipation (10). 5. Measure the average power dissipation of this circuit, by adding the right command in the “Additional spice commands” of the S-Edit Spice-Simulation dialog window. 6. How does this power dissipation figure compare to the one of the axon-hillock circuit? What is the reason for this behavior? Explain this in the report. Congratulations, you have done your first transient simulation in SPICE. Perhaps now you can understand why people both love and hate simulators. Never forget that this is only a simulation, and even the best transistor models don’t capture everything that is important for analog (or even digital) circuit performance. Neuromorphic Engineering II Lab 2, Spring 2013 2.5 5 What we expect How to use basic SPICE commands to simulate circuits. What are the limitations of circuit simulators? What characteristics of circuits do they not model? What makes a low-power silicon neuron so low-power? 2.6 Next Week Layout using LEDIT.