Magic Tutorial #2: Basic Painting and Selection John Ousterhout Computer Science Division Electrical Engineering and Computer Sciences University of California Berkeley, CA 94720 (Updated by others, too.) This tutorial corresponds to Magic version 6.
Tutorials to read first:
Commands introduced in this tutorial:
Macros introduced in this tutorial:
In Magic, a circuit layout is a hierarchical collection of cells. Each cell contains three things: colored shapes, called paint, that define the circuit's structure; textual labels attached to the paint; and subcells, which are instances of other cells. The paint is what determines the eventual function of the VLSI circuit. Labels and subcells are a convenience for you in managing the layout and provide a way of communicating information between various synthesis and analysis tools. This tutorial explains how to create and edit paint and labels in simple single-cell designs, using the basic painting commands. ``Magic Tutorial #3: Advanced Painting (Wiring and Plowing)'' describes some more advanced features for manipulating paint. For information on how to build up cell hierarchies, see ``Magic Tutorial #4: Cell Hierarchies''.
Enter Magic to edit the cell tut2a (type magic tut2a to the Unix shell; follow the directions in ``Tutorial #1: Getting Started'' if you have any problems with this). The tut2a cell is a sort of palette: it shows a splotch of each of several paint layers and gives the names that Magic uses for the layers.
The two basic layout operations are painting and erasing. They can be invoked using the :paint and :erase long commands, or using the buttons. The easiest way to paint and erase is with the mouse buttons. To paint, position the box over the area you'd like to paint, then move the cursor over a color and click the middle mouse button. To erase everything in an area, place the box over the area, move the cursor over a blank spot, and click the middle mouse button. Try painting and erasing various colors. If the screen gets totally messed up, you can always exit Magic and restart it. While you're painting, white dots may occasionally appear and disappear. These are design rule violations detected by Magic, and will be explained in ``Magic Tutorial #6: Design Rule Checking''. You can ignore them for now.
It's completely legal to paint one layer on top of another. When this happens, one of three things may occur. In some cases, the layers are independent, so what you'll see is a combination of the two, as if each were a transparent colored foil. This happens, for example, if you paint metal1 (blue) on top of polysilicon (red). In other cases, when you paint one layer on top of another you'll get something different from either of the two original layers. For example, painting poly on top of ndiff produces ntransistor (try this). In still other cases the new layer replaces the old one: this happens, for example, if you paint a pcontact on top of ntransistor. Try painting different layers on top of each other to see what happens. The meaning of the various layers is discussed in more detail in Section 11 below.
There is a second way of erasing paint that allows you to erase some layers without affecting others. This is the macro ^D (control-D, for ``Delete paint''). To use it, position the box over the area to be erased, then move the crosshair over a splotch of paint containing the layer(s) you'd like to erase. Type ^D key on the text keyboard: the colors underneath the cursor will be erased from the area underneath the box, but no other layers will be affected. Experiment around with the ^D macro to try different combinations of paints and erases. If the cursor is over empty space then the ^D macro is equivalent to the middle mouse button: it erases everything.
You can also paint and erase using the long commands
:paint layers :erase layers
In each of these commands layers is one or more layer names separated by commas (you can also use spaces for separators, but only if you enclose the entire list in double-quotes). Any layer can be abbreviated as long as the abbreviation is unambiguous. For example, :paint poly,metal1 will paint the polysilicon and metal1 layers. The macro ^D is predefined by Magic to be :erase $ ($ is a pseudo-layer that means ``all layers underneath the cursor'').
There are probably going to be times when you'll do things that you'll later wish you hadn't. Fortunately, Magic has an undo facility that you can use to restore things after you've made mistakes. The command
:undo
(or, alternatively, the macro u) will undo the effects of the last command you invoked. If you made a mistake several commands back, you can type :undo several times to undo successive commands. However, there is a limit to all this: Magic only remembers how to undo the last ten or so commands. If you undo something and then decide you wanted it after all, you can undo the undo with the command
:redo
(U is a macro for this command). Try making a few paints and erases, then use :undo and :redo to work backwards and forwards through the changes you made.
Once you have painted a piece of layout, there are several commands you can invoke to modify the layout. Many of them are based on the selection: you select one or more pieces of the design, and then perform operations such as copying, deletion, and rotation on the selected things. To see how the selection works, load cell tut2b. You can do this by typing :load tut2b if you're still in Magic, or by starting up Magic with the shell command magic tut2b.
The first thing to do is to learn how to select. Move the cursor over the upper portion of the L-shaped blue area in tut2b, and type s, which is a macro for :select. The box will jump over to cover the vertical part of the ``L''. This operation selected a chunk of material. Move the box away from the chunk, and you'll see that a thin white outline is left around the chunk to show that it's selected. Now move the cursor over the vertical red bar on the right of the cell and type s. The box will move over that bar, and the selection highlighting will disappear from the blue area.
If you type s several times without moving the cursor, each command selects a slightly larger piece of material. Move the cursor back over the top of the blue ``L'', and type s three times without moving the cursor. The first s selects a chunk (a rectangular region all of the same type of material). The second s selects a region (all of the blue material in the region underneath the cursor, rectangular or not). The third s selects a net (all of the material that is electrically connected to the original chunk; this includes the blue metal, the red polysilicon, and the contact that connects them).
The macro S (short for :select more) is just like s except that it adds on to the selection, rather than replacing it. Move the cursor over the vertical red bar on the right and type S to see how this works. You can also type S multiple times to add regions and nets to the selection.
If you accidentally type s or S when the cursor is over space, you'll select a cell (tut2b in this case). You can just undo this for now. Cell selection will be discussed in ``Magic Tutorial #4: Cell Hierarchies''.
You can also select material by area: place the box around the material you'd like to select and type a (short for :select area). This will select all of the material underneath the box. You can use the macro A to add material to the selection by area, and you can use the long command
:select [more] area layers
to select only material on certain layers. Place the box around everything in tut2b and type :select area metal1 followed by :select more area poly.
If you'd like to clear out the selection without modifying any of the selected material, you can use the command
:select clear
or type the macro C. You can clear out just a portion of the selection by typing :select less or :select less area layers; the former deselects paint in the order that :select selects paint, while the latter deselects paint under the box (just as :select area selects paint under the box). For a synopsis of all the options to the :select command, type
:select help
Once you've made a selection, there are a number of operations you can perform on it:
:delete :move [direction [distance]] :stretch [direction [distance]] :copy :upsidedown :sideways :clockwise [degrees]
The :delete command deletes everything that's selected. Watch out: :delete is different from :erase, which erases paint from the area underneath the box. Select the red bar on the right in tut2b and type d, which is a macro for :delete. Undo the deletion with the u macro.
The :move command picks up both the box and the selection and moves them so that the lower-left corner of the box is at the cursor location. Select the red bar on the right and move it so that it falls on top of the vertical part of the blue ``L''. You can use t (``translate'') as a macro for :move. Practice moving various things around the screen. The command :copy and its macro c are just like :move except that a copy of the selection is left behind at the original position.
There is also a longer form of the :move command that you can use to move the selection a precise amount. For example, :move up 10 will move the selection (and the box) up 10 units. The direction argument can be any direction like left, south, down, etc. See the Magic manual page for a complete list of the legal directions. The macros q, w, e, and r are defined to move the selection left, down, up, and right (respectively) by one unit.
The :stretch command is similar to :move except that it stretches and erases as it moves. :stretch does not operate diagonally, so if you use the cursor to indicate where to stretch to, Magic will either stretch up, down, left, or right, whichever is closest. The :stretch command moves the selection and also does two additional things. First, for each piece of paint that moves, :stretch will erase that layer from the region that the paint passes through as it moves, in order to clear material out of its way. Second, if the back edge of a piece of selected paint touches non-selected material, one of the two pieces of paint is stretched to maintain the connection. The macros Q, W, E, and R just like the macros q, etc. described above for :move. The macro T is predefined to :stretch. To see how stretching works, select the horizontal piece of the green wire in tut2b and type W, then E. Stretching only worries about material in front of and behind the selection; it ignores material to the sides (try the Q and R macros to see). You can use plowing (described in Tutorial #3) if this is a problem.
The command :upsidedown will flip the selection upside down, and :sideways flips the selection sideways. Both commands leave the selection so it occupies the same total area as before, but with the contents flipped. The command :clockwise will rotate the selection clockwise, leaving the lower-left corner of the new selection at the same place as the lower-left corner of the old selection. Degrees must be a multiple of 90, and defaults to 90.
At this point you know enough to do quite a bit of damage to the tut2b cell. Experiment with the selection commands. Remember that you can use :undo to back out of trouble.
Labels are pieces of text attached to the paint of a cell. They are used to provide information to other tools that will process the circuit. Most labels are node names: they provide an easy way of referring to nodes in tools such as routers, simulators, and timing analyzers. Labels may also be used for other purposes: for example, some labels are treated as attributes that give Crystal, the timing analyzer, information about the direction of signal flow through transistors.
Load the cell tut2c and place a cross in the middle of the red chunk (to make a cross, position the lower-left corner of the box with the left button and then click the right button to place the upper-right corner on top of the lower-left corner). Then type type the command :label test. A new label will appear at the position of the box. The complete syntax of the :label command is
:label [text [position [layer]]]
Text must be supplied, but the other arguments can be defaulted. If text has any spaces in it, then it must be enclosed in double quotes. Position tells where the text should be displayed, relative to the point of the label. It may be any of north, south, east, west, top, bottom, left, right, up, down, center, northeast, ne, southeast, se, southwest, sw, northwest, nw. For example, if ne is given, the text will be displayed above and to the right of the label point. If no position is given, Magic will pick a position for you. Layer tells which paint layer to attach the label to. If layer covers the entire area of the label, then the label will be associated with the particular layer. If layer is omitted, or if it doesn't cover the label's area, Magic initially associates the label with the ``space'' layer, then checks to see if there's a layer that covers the whole area. If there is, Magic moves the label to that layer. It is generally a bad idea to place labels at points where there are several paint layers, since it will be hard to tell which layer the label is attached to. As you edit, Magic will ensure that labels are only attached to layers that exist everywhere under the label. To see how this works, paint the layer pdiff (brown) over the label you just created: the label will switch layers. Finally, erase poly over the area, and the label will move again.
Although many labels are point labels, this need not be the case. You can label any rectangular area by setting the box to that area before invoking the label command. This feature is used for labelling terminals for the router (see below), and for labelling tiles used by Mpack, the tile packing program. Tut2c has examples of point, line, and rectangular labels.
All of the selection commands apply to labels as well as paint. Whenever you select paint, the labels attached to that paint will also be selected. Selected labels are highlighted in white. Select some of the chunks of paint in tut2c to see how the labels are selected too. When you use area selection, labels will only be selected if they are completely contained in the area being selected. If you'd like to select just a label without any paint, make the box into a cross and put the cross on the label: s and S will select just the label.
There are several ways to erase a label. One way is to select and then delete it. Another way is to erase the paint that the label is attached to. If the paint is erased all around the label, then Magic will delete the label too. Try attaching a label to a red area, then paint blue over the red. If you erase blue the label stays (since it's attached to red), but if you erase the red then the label is deleted.
You can also erase labels using the :erase command and the pseudo-layer labels. The command
:erase labels
will erase all labels that lie completely within the area of the box. Finally, you can erase a label by making the box into a cross on top of the label, then clicking the middle button with the cursor over empty space. Technically, this will erase all paint layers and labels too. However, since the box has zero area, erasing paint has no effect: only the labels are erased.
When creating labels, Magic will permit you to use absolutely any text whatsoever. However, many other tools, and even parts of Magic, expect label names to observe certain conventions. Except for the special cases described below, labels shouldn't contain any of the characters ``/$@!^''. Spaces, control characters, or parentheses within labels are probably a bad idea too. Many of the programs that process Magic output have their own restrictions on label names, so you should find out about the restrictions that apply at your site. Most labels are node names: each one gives a unique identification to a set of things that are electrically connected. There are two kinds of node names, local and global. Any label that ends in ``!'' is treated as a global node name; it will be assumed that all nodes by this name, anywere in any cell in a layout, are electrically connected. The most common global names are Vdd! and GND!, the power rails. You should always use these names exactly, since many other tools require them. Nobody knows why ``GND!'' is all in capital letters and ``Vdd!'' isn't.
Any label that does not end in ``!'' or any of the other special characters discussed below is a local node name. It refers to a node within that particular cell. Local node names should be unique within the cell: there shouldn't be two electrically distinct nodes with the same name. On the other hand, it is perfectly legal, and sometimes advantageous, to give more than one name to the same node. It is also legal to use the same local node name in different cells: the tools will be able to distinguish between them and will not assume that they are electrically connected.
The only other labels currently understood by the tools are attributes. Attributes are pieces of text associated with a particular piece of the circuit: they are not node names, and need not be unique. For example, an attribute might identify a node as a chip input, or it might identify a transistor terminal as the source of information for that transistor. Any label whose last character is ``@'', ``$'', or ``^'' is an attribute. There are three different kinds of attributes. Node attributes are those ending with ``@''; they are associated with particular nodes. Transistor source/drain attributes are those ending in ``$''; they are associated with particular terminals of a transistor. A source or drain attribute must be attached to the channel region of the transistor and must fall exactly on the source or drain edge of the transistor. The third kind of attribute is a transistor gate attribute. It ends in ``^'' and is attached to the channel region of the transistor. To see examples of attributes and node names, edit the cell tut2d in Magic.
Special conventions apply to labels for routing terminals. The standard Magic router (invoked by :route) ignores all labels except for those on the edges of cells. (This restriction does not apply to the gate-array router, Garoute, or to the interactive router, Iroute). If you expect to use the standard router to connect to a particular node, you should place the label for that node on its outermost edge. The label should not be a point label, but should instead be a horizontal or vertical line covering the entire edge of the wire. The router will choose a connection point somewhere along the label. A good rule of thumb is to label all nodes that enter or leave the cell in this way. For more details on how labels are used by the standard router, see ``Magic Tutorial #7: Netlists and Routing''. Other labeling conventions are used by the Garouter and Irouter, consult their respective tutorials for details.
Magic provides a variety of ways to save your cells on disk. Normally, things are saved in a special Magic format. Each cell is a separate file, and the name of the file is just the name of the cell with .mag appended. For example, the cell tut2a is saved in file tut2a.mag. To save cells on disk, invoke the command
:writeall
This command will run through each of the cells that you have modified in this editing session, and ask you what to do with the cell. Normally, you'll type write, or just hit the return key, in which case the cell will be written back to the disk file from which it was read (if this is a new cell, then you'll be asked for a name for the cell). If you type autowrite, then Magic will write out all the cells that have changed without asking you what to do on a cell-by-cell basis. Flush will cause Magic to delete its internal copy of the cell and reload the cell from the disk copy, thereby expunging all edits that you've made. Skip will pass on to the next cell without writing this cell (but Magic still remembers that it has changed, so the next time you invoke :writeall Magic will ask about this cell again). Abort will stop the command immediately without writing or checking any more cells.
IMPORTANT NOTE: Unlike vi and other text editors, Magic doesn't keep checkpoint files. This means that if the system should crash in the middle of a session, you'll lose all changes since the last time you wrote out cells. It's a good idea to save your cells frequently during long editing sessions.
You can also save the cell you're currently editing with the command
:save name
This command will append ``.mag'' to name and save the cell you are editing in that location. If you don't provide a name, Magic will use the cell's name (plus the ``.mag'' extension) as the file name, and it will prompt you for a name if the cell hasn't yet been named.
Once a cell has been saved on disk you can edit it by invoking Magic with the command
magic name
where name is the same name you used to save the cell (no ``.mag'' extension).
Magic can also read and write files in CIF and Calma Stream formats. See ``Magic Tutorial #9: Format Conversion for CIF and Calma'' for details.
Magic can generate hardcopy plots of layouts in four ways: versatec (black-and-white or color), gremlin and pixels (a generalized pixel-file that can be massaged in many ways). The first style is for printers like the black-and-white Versatec family: for these, Magic will output a raster file and spool the file for printing. To plot part of your design, place the box around the part you'd like to plot and type
:plot versatec [width [layers]]
This will generate a plot of the area of the box. Everything visible underneath the box will appear in more-or-less the same way in the plot. Width specifies how wide the plot will be, in inches. Magic will scale the plot so that the area of the box comes out this wide. The default for width is the width of the plotter (if width is larger than the plotter width, it's reduced to the plotter width). If layers is given, it specifies exactly what information is to be plotted. Only those layers will appear in the plot. The special ``layer'' labels will enable label plotting.
The second form is for driving printers like color Versatecs. It is enabled by setting the color plot parameter to true. A table of stipples for the primary colors (black, cyan, magenta abd yellow) is given in the technology file. When the plot command is given, four rasters (one for each of the colors) are generated, separated with the proper control sequences for the printer. Otherwise, operation is exactly as for the black-and-white case.
The third form of plotting is for generating Gremlin-format files, which can then be edited with the Gremlin drawing system or included in documents processed by Grn and Ditroff. The command to get Gremlin files is
:plot gremlin file [layers]
It will generate a Gremlin-format file in file that describes everything underneath the box. If layers is specified, it indicates which layers are to appear in the file; otherwise everything visible on the screen is output. The Gremlin file is output without any particular scale; use the width or height commands in Grn to scale the plot when it's printed. You should use the mg stipples when printing Magic Gremlin plots; these will produce the same stipple patterns as :plot versatec.
Finally, the ``pixels'' style of plotting generates a file of pixel values for the region to be plotted. This can be useful for input to other image tools, or for generation of slides and viewgraphs for presentations. The file consists of a sequence of bytes, three for each pixel, written from left to right and top to bottom. Each three bytes represent the red, green and blue values used to display the pixel. Thus, if the upper-left-most pixel were to be red, the first three bytes of the file would have values of 255, 0 and 0.
The resolution of the generated file is normally 512, but can be controlled by setting the plot parameter pixWidth. It must be a multiple of 8; Magic will round up if an inappropriate value is entered. The height of the file is determined by the shape of the box. In any case, the actual resolution of the file is appended to the file name. For example, plotting a square region, 2048 pixels across, will result in a file named something like ``magicPlot1234a-2048-2048''.
There are several plotting parameters used internally to Magic, such as the width of the Versatec printer and the number of dots per inch on the Versatec printer. You can modify most of these to work with different printers. For details, read about the various :plot command options in the man page.
There are several additional commands that you will probably find useful once you start working on real cells. The command
:grid [spacing] :grid xSpacing ySpacing :grid xSpacing ySpacing xOrigin yOrigin :grid off
will display a grid over your layout. Initially, the grid has a one-unit spacing. Typing :grid with no arguments will toggle the grid on and off. If a single numerical argument is given, the grid will be turned on, and the grid lines will be spacing units apart. The macro g provides a short form for :grid and G is short for :grid 2. If you provide two arguments to :grid, they are the x- and y-spacings, which may be different. If you provide four arguments, the last two specify a reference point through which horizontal and vertical grid lines pass; the default is to use (0,0) as the grid origin. The command :grid off always turns the grid off, regardless of whether or not is was previously on. When the grid is on, a small black box is displayed to mark the (0,0) coordinate of the cell you're editing.
If you want to create a cell that doesn't fit on the screen, you'll need to know how to change the screen view. This can be done with three commands:
:zoom factor :findbox [zoom] :view
If factor is given to the zoom command, it is a zoom-out factor. For example, the command :zoom 2 will change the view so that there are twice as many units across the screen as there used to be (Z is a macro for this). The new view will have the same center as the old one. The command :zoom .5 will increase the magnification so that only half as much of the circuit is visible.
The :findbox command is used to change the view according to the box. The command alone just moves the view (without changing the scale factor) so that the box is in the center of the screen. If the zoom argument is given then the magnification is changed too, so that the area of the box nearly fills the screen. z is a macro for :findbox zoom and B is a macro for :findbox.
The command :view resets the view so that the entire cell is visible in the window. It comes in handy if you get lost in a big layout. The macro v is equivalent to :view.
The command :box prints out the size and location of the box in case you'd like to measure something in your layout. The macro b is predefined to :box. The :box command can also be used to set the box to a particular location, height, or width. See the man page for details.
The command
:what
will print out information about what's selected. This may be helpful if you're not sure what layer a particular piece of material is, or what layer a particular label is attached to.
If you forget what a macro means, you can invoke the command
:macro [char]
This command will print out the long command that's associated with the macro char. If you omit char, Magic will print out all of the macro associations. The command
:macro char command
We set up char to be a macro for command, replacing the old char macro if there was one. If command contains any spaces then it must be enclosed in double-quotes. To see how this works, type the command :macro 1 "echo You just typed the 1 key.", then type the 1 key.
One of the macros, ``.'', has special meaning in Magic. This macro is always defined by the system to be the last long command you typed. Whenever you'd like to repeat a long command, all you have to do is use the dot macro.
The paint layers available in Magic are different from those that you may be used to in Caesar and other systems because they don't correspond exactly to the masks used in fabrication. We call them abstract layers because they correspond to constructs such as wires and contacts, rather than mask layers. We also call them logs because they look like sticks except that the geometry is drawn fully fleshed instead of as lines. In Magic there is one paint layer for each kind of conducting material (polysilicon, ndiffusion, metal1, etc.), plus one additional paint layer for each kind of transistor (ntransistor, ptransistor, etc.), and, finally, one further paint layer for each kind of contact (pcontact, ndcontact, m2contact, etc.) Each layer has one or more names that are used to refer to that layer in commands. To find out the layers available in the current technology, type the command
:layers
In addition to the mask layers, there are a few pseudo-layers that are valid in all technologies; these are listed in Table I. Each Magic technology also has a technology manual describing the features of that technology, such as design rules, routing layers, CIF styles, etc. If you haven't seen any of the technology manuals yet, this is a good time to take a look at the one for your process.
+-----------------------------------------+
|errors (design-rule violations) |
|labels |
|subcells |
|* (all mask layers) |
|$ (all mask layers visible under cursor) |
+-----------------------------------------+
If you're used to designing with mask layers (e.g. you've been reading the Mead-Conway book), Magic's log style will take some getting used to. One of the reasons for logs is to save you work. In Magic you don't draw implants, wells, buried windows, or contact via holes. Instead, you draw the primary conducting layers and paint some of their overlaps with special types such as n-transistor or polysilicon contact. For transistors, you draw only the actual area of the transistor channel. Magic will generate the polysilicon and diffusion, plus any necessary implants, when it creates a CIF file. For contacts, you paint the contact layer in the area of overlap between the conducting layers. Magic will generate each of the constituent mask layers plus vias and buried windows when it writes the CIF file. Figure 1 shows a simple cell drawn with both mask layers (as in Caesar) and with logs (as in Magic). If you're curious about what the masks will look like for a particular layout, you can use the :cif see command to view the mask information.
An advantage of the logs used in Magic is that they simplify the design rules. Most of the formation rules (e.g. contact structure) go away, since Magic automatically generates correctly-formed structures when it writes CIF. All that are left are minimum size and spacing rules, and Magic's abstract layers result in fewer of these than there would be otherwise. This helps to make Magic's built-in design rule checker very fast (see ``Magic Tutorial #6: Design Rule Checking''), and is one of the reasons plowing is possible.