Create from Image

Drop a picture of a schematic or a layout and get its circuit as a schematic, or a layout's copper as a layout — what was read drawn over the picture, and every guess one click from being corrected.

What it does

A picture of a circuit — a datasheet's application circuit, another tool's screenshot, a board viewer's screenshot, a die plot, a figure in an application note — holds the circuit, but not in a form a simulator can use. Create from Image reads it:

• Create Schematic from Image, on a picture of a schematic, reads its wires, symbols, designators and values and draws them as a schematic of native components, laid out as the picture lays them out — see Schematic pictures.
• Create Layout from Image traces a layout picture's copper into a layout: filled shapes on your technology's layers, drill holes as vias, sitting exactly on the picture. A picture of a board makes a layout only, whichever command opened it; to get its schematic, run Create Schematic from Artwork on the traced layout.
• Create Touchstone from Plot, on pictures of S-parameter plots — dB and phase against frequency, a Smith chart — digitises their curves into a Touchstone file — see Plots.

Like reading artwork, it is a best attempt that you then check. Everything it guessed is shown on the picture and said in the report.

Ways in

Wherever there is a picture, the command is one gesture away — and it is only offered where it can work.

A placed picture whose file cannot be found offers none of these rows — point it at its file first (on a layout, Resolve Path… on the same menu). A locked picture is offered like any other: locking stops it moving, not being read.

Dropping a picture onto a canvas still places it as a bitmap, and dropping one onto the Project panel still adds it as a Known File; the right-click rows above are the next step. Dropping one onto the dialog reads it.

The dialog

The dialog opens empty, as one drop zone. Drop a picture on it, paste one (Cmd/Ctrl+V), or Browse…. It reads PNG, JPEG, BMP, GIF (the first frame) and WebP; a file that is not a picture is refused in the status line, and nothing else changes.

With a picture, the top of the dialog is the picture and what to do with it; below it is the body the artwork dialog has — target, options, parts table, report and Create — with the picture itself taking the left two-thirds. The buttons along the top are glyphs, each named in its tooltip: Replace and Paste read another picture in its place, and Help, at the top right, opens this page at the section for what is on screen.

What kind of picture it is

Read as says what kind of picture it is. Auto decides from the picture — how much of it is ink, whether the ink is filled or stroked, how many colours it has, whether its lines run straight — and its tooltip says what it decided, with a dot beside it for how sure it is: green sure, amber fairly sure, red a guess. Hover over it to see what it measured. Schematic, Layout and Plot override it; a picture Auto reads wrongly is read your way after one click.

A sparse board of one trace width can look to Auto like a schematic drawing. If it does, choose Layout.

Design ▸ Create ▸ Layout from Image… opens on Layout, and every picture you drop, paste or browse to while it is open is read as a layout straight away, without Auto deciding first. Choose Auto to have it decide.

A photograph of a board reads as Auto (board photo), with a less certain dot than a drawing gets: it is read as a layout, and is best set square on its corners first — see Photographs of boards.

What to make of it

Make is Schematic or Layout. A schematic picture makes a schematic only: draw the schematic first, then use Update Layout from Schematic. Layout is disabled for it, and its tooltip says so. A picture of a board makes a layout only, so Schematic is disabled for it.

The Target is a new cell, named after the picture (a pasted picture is pasted_image_1, _2, …). A name that is a cell this command made relabels Create as Replace, and a history checkpoint is taken before anything is replaced. A cell this command did not make is never replaced.

Making a schematic from a schematic picture also shows Arrange and Crossings connect (Schematic pictures), the frequency sweep and the digits values are written with.

The picture, and what was read

What was read is drawn over the picture, each kind toggled from the legend at the top left:

Click a part on the picture to select its row in the parts table; select a row, or an anchor in the report, to highlight its part on the picture.

Correcting a layout picture by hand

The colours say where the copper is, not what every piece of it is. Its pieces can be settled by hand before Create, with or without a scale. Each choice is drawn dashed over the picture, is kept through every re-read, and is undone with Cmd/Ctrl+Z on the picture; Clear corrections forgets them all. Every choice is also a row of the Corrections list under Layers.

Click a hole to say what it is:

And every one like it (on) answers every hole of the same diameter, within 10 %, the same way: seventy vias are one click. A technology with no non-plated hole layer is never changed: the new cell gets a copy of it with one added, Non-plated hole, spanning the board, written beside the cell as <cell>.ctech, and the report says so.

Click a copper region: Copper on a conductor chosen beside it, or Not copper — a part's body the colours read as copper; nothing is written inside it, on any layer. A traced region's shape is not changed here: to reshape one, make it Not copper and draw a Copper fill in its place, or reshape it after Create in the layout editor, with Keep the source picture under the result putting the picture under it.

Draw with the tools beside the Zoom to Fit button, each armed until Escape. The armed tool's choice — the conductor, or the via's drill — appears under the tools; nothing in the row of tools moves when one is armed.

Point at anything drawn over the picture to see what it is, in one line: Via ⌀0.6 mm, Opening ⌀2.4 mm — a hole kept open, Painted copper on Top Copper, Board outline. A hole is read as one only when it is round: a square or rounded-square hole is a pad's, and the board inside a drawn outline is never a hole, whatever the scale. Openings wider than any via are counted in the report: click one to make it a via or a non-plated hole.

The Drill picker offers:

Edit a correction. Click one on the picture, or its row in the list, to select it. Delete, or the row's ✕, removes it. A selected copper fill, area cleared or outline shows its corners: drag one to move it, double-click an edge to add one, click one and press Delete to remove it (a shape keeps at least three). A selected via placed by hand is dragged to a new place, and the Drill picker gives it another size — with And every one like it on, every via placed by hand at its size. A copper fill's conductor is chosen on its row. Each edit is one undo step.

Remove ▾ clears one kind at once, each one undo step, each item saying how many it removes (a kind with none is greyed): Vias read from the picture (their holes stay openings), Vias placed by hand, Copper fills drawn, Paint fills, Clears drawn, The outline drawn, and Copper read on a conductor — everything traced on it, so only the copper you draw there is written.

The board outline — drawn, set by Board corners… ▸ Use as the board outline, or read from the picture — stays drawn above the copper in the accent colour after the tool is put down, the picture outside it dimmed. Once one is drawn, only the copper inside it is read.

A choice that lands on nothing — a via off the copper, a hole where there is none — is named in the report and not applied. The report counts the rest: 4 corrections: 2 vias added, 1 hole not plated, 1 copper fill on Top Copper.

Schematic pictures

Not available in this version. Design ▸ Create ▸ Schematic from Image… and the dialog's Schematic buttons are greyed while the schematic reader is improved, and a picture read as a schematic drawing is traced as a layout. What follows describes the reader for when it returns.

A picture of a schematic is read as its wires, its symbols and its words, and drawn as a schematic of native components.

What is recognised. Resistors, inductors and capacitors in both the US and the IEC drawings — variable resistors, looped coils and curved plates among them — diodes, transformers, grounds, ports and terminals (terminal circles, filled dots, coax connectors), transmission lines and stubs. A transistor in its circle is circuitRF's own BJT or FET; an AC source circle is a P1Tone, and a DC source, a battery or a + supply circle a Vdc. An amplifier triangle is the System Amp, and filter, attenuator, circulator, switch, coupler and balun symbols are their System components, each on default parameters. An IC box, and any other box with three or more pins, is an S-parameter block (SnP) with one port per signal pin, in the order of the pin numbers printed beside them, and its ground pins as its reference — an S2P drawn as a box with pins 1, 2 and Ref reads as an S2P — ready for its Touchstone file. A two-pin symbol nothing matches is shown as ? in the parts table and drawn as a 0 Ω part named ?1, ?2, …, so the circuit stays as drawn until you say what it is. Current arrows, dimension arrows, dashed boxes and dashed continuation lines are notes on the drawing, never wires.

How wires are read. A T-junction always connects. Where two wires cross, Crossings connect decides: With a dot (the default) connects a crossing only where a junction dot is drawn; Never and Always do what they say. A hop never connects. Wires with the same net label, or the same supply name, are one net. A wire end that reaches nothing is shown in amber.

Correcting a connection. Click a crossing on the picture to connect or disconnect that one; drag a wire end that reaches nothing onto a part's pin to connect it. Each correction is kept through every later reading, and Clear corrections forgets them all. Nothing else is edited on the picture — the schematic editor is where a drawing is fixed.

Values. Designators, values, net names and port names are read; prose is not. A value may be written alone (10pF, 4R7, 2n2) or with its parameter's name and spaces (L = 5 nH, C = 0.5 pF); a value printed on the line under a part's designator is that part's, wherever the label stands. A port's Z = 75 Ohm under its designator sets its impedance. A value that cannot be read becomes a global variable named after its part — C1_C, L2_L — with a tune entry, starting at a value that leaves the circuit as if the part were not there, so the schematic simulates at once and the Tuning panel lists what is missing. A transmission line takes its Z and E (or its length) from the text beside it, and each one not read is a variable in the same way. The parts table lists the parts with their position on the picture in pixels; edit a kind or a value there exactly as for artwork.

Ports. Port and terminal symbols are ports, and so are net names that name a port (RFIN, OUT, P1); nothing else is — a wire running off the edge of the picture stays a wire. A schematic picture with no port at all is still made: it is written without its S-parameter analysis, and the report says No port found: add Port components to simulate. A supply net, such as VCC, has no source in the circuit and is left as a named net.

Arrange. As drawn puts each part where the picture has it, turned as it is drawn, and draws the picture's own wires — every corner, junction and loose end — snapped to the schematic grid, with a ground symbol where the picture draws one, a net label only where the picture writes a net's name, and the picture under it. A screenshot of a schematic editor is drawn at the scale of its own grid, so its wires land on the schematic's grid lines and the picture lines up under the drawing. A part that would land on another is placed beside the drawing instead, and the report names it. Tidy draws the same circuit in circuitRF's own arrangement, signal path left to right, without the picture under it — it would not line up.

Readings. A schematic picture is read up to four ways — at the scales measured from it; enlarged once more, for a small, low-resolution picture; with every crossing joined, for a drawing that joins its wires without dots; and with bold type read as drawn — and the most plausible reading is kept: the one that reads most of the ink as wires, parts and words, finds a part in each symbol, and leaves no net going nowhere. Reading lists them, each with its plausibility; pick another to see it at once, and Create makes the one shown. A way of reading that cannot change anything on the picture is not tried, and a large picture is read once. Stating a Pen, Text height or Symbol scale, or correcting a connection, reads it the one way.

Making a layout of it. A schematic picture makes a schematic only. To get a layout, create the schematic, then use Update Layout from Schematic.

Reading a circuit from a datasheet or a textbook

Schematic from Image… reads a screenshot of a datasheet or textbook circuit, so it is worth knowing what it reads well and what it leaves for you.

What pastes well. An application or test circuit cropped from a datasheet, at screen resolution or on a Retina display, and a textbook or web schematic drawn by a program. Hairline pens, coloured text and filled amplifier triangles, an IC box with its pin numbers and pin names, coax connectors, small circles marking pins on the leads, bold labels, serif italic designators with subscripts (R₁, C_ADJ), zig-zag resistors and named open wire ends are all read.

What to expect to fix.

• An IC's Touchstone file. An IC box is an S-parameter block with no file; set its File before simulating, or the run stops at it.
• An amplifier's gain. The Amp stands on 20 dB, a global variable with a tune entry; set the part's real gain.
• A value the picture did not print, or printed where it could not be read, is a global variable with a tune entry.
• A part of no known kind is a 0 Ω part named ?n: set its kind in the dialog's parts table, or replace it in the schematic.
• A supply — an arrow or a terminal named VCC, +5 V, Vs — is a named net with no source; add a Vdc if the analysis needs the bias.

What does not read yet. A coil drawn in a hairline pen can come apart into several ? parts, and two capacitors drawn side by side can read as one. An IC box divided by small pin-number boxes can read as several blocks. A solid box drawn round a part or a module — a test fixture, a bias tee, a frame round the device under test — is read as one block with what it holds. A transistor drawn without its circle, and a hatched or dotted zig-zag, are not read. A part marked not fitted only by its colour is read like any other. An op-amp circuit or a dense evaluation board reads its parts, but most of their connections need redrawing. A complex picture takes up to about twenty seconds to read.

Scale

A picture has no millimetres in it, and a wrong scale is wrong by a factor while looking entirely right — so the scale is never a silent guess. The Scale row shows the length of one pixel and what it was taken from, and the combo beside it lists every piece of evidence found:

The distance needs its unit — 1.6 mm, 62 mil, 500 um. A bare number is refused where you typed it: it could be any of them. Beside the value, and on the picture with the scale buttons beside it, the row says how big the board is at that scale — its outline once drawn, else the crop, else the whole picture: Board 17 mm × 17 mm. If that is not the board's real size, the scale is wrong: a Line impedance… taken with a technology whose stackup is not the board's gives a board several times too small or too large. Then the resolution line says how finely the picture can be read: 1 px = 20 µm; a 451 µm trace is read to ± 2 %.

With no usable evidence, the row is highlighted, the picture asks Set the scale: with Two points…, Board corners… and Line impedance… as buttons beside it, and Create waits.

Photographs of boards

A phone photo of a board is taken at an angle, in uneven light, with most of the copper under the solder mask. Choose Board corners… in the Scale combo: four handles appear on the photo as taken. Drag them onto the board's corners, clockwise from its top-left — or onto any rectangle whose size you know, such as a part or a connector's outline — type its size in the box between the handles (50 mm × 30 mm, each side with its unit), and press Set. The first handle to the second is the width. Board corners… works on a drawing as well as a photo: on a datasheet's board drawing, the board's outline and its stated size set the scale in one step. Use as the board outline, under the size and on by default, writes the rectangle as the board's outline.

The photo is then read square: the picture shows the rectified board, the rectangle's size is the scale, and the light is evened out across it. Its pixels are read as six classes, each a row under Layers in a fixed colour, so a preset saved from one board photo applies to the next:

Glare — a white glint — is filled from the surface around it, so a line running under it stays one line. A part's body hides its pads: each one is counted in the report and not read. The report also gives how far each corner of the rectangle lies from the corner the photo shows there, in µm; a large one means a handle is off its corner. Read as taken, beside the size, puts the photo back as it was.

Widths read from a photo are good enough to start a schematic from, not to trust for impedance: a line's Z0 from its photographed width is an estimate.

Plots

A datasheet often shows a part's S-parameters as plots and ships no file. A picture of a plot — dB or phase against frequency, a Smith chart, a polar chart — is read as Plot, and Make is Touchstone or CSV.

Read is Automatic or Manual. Automatic finds each plot, reads its tick labels and traces its curves, and you correct what it read; the two Touchstone rows open in it. Manual reads nothing off the picture: you draw the plot area, type its scale and place each curve's points, so it works on any picture — see Digitise by hand. Digitise Image… opens in Manual, or in Automatic if that is what you chose there last. Switching keeps the picture.

Several pictures make one file. A datasheet's S21 and S11 are often separate figures: Add picture… under Pictures, or drop or paste another, adds it to the same session; ✕ takes one out. Plots and traces are numbered across all of them — plot 1 is the first picture's first plot, and traces are T1, T2 and so on. Replace… starts over: the session's pictures are set aside and the one chosen is read as picture 1.

On the picture, each plot is boxed and numbered, its tick labels green where the axis used them and amber where they were not read or were outvoted, and each trace is drawn over its curve in its own colour and labelled with what it was read as (S21 dB). Markers are ringed, a gap bridged over a label or another curve is dashed, and a crossing of two curves of one colour that could not be told apart is amber. Under Plots, clear a plot's box to leave it out: it is drawn and not read.

A picture holding several plots, or a legend read as a trace, is answered with a box dragged on the picture, using the two tools beside Zoom to Fit. Read only here crops the picture to the box and reads only that part; Restore whole picture brings the rest back. Ignore here reads nothing inside the box — a legend or a note drawn inside a plot — and the picture is read again. The box stays on the picture, hatched; click it and press Delete to remove it. A box outside every plot is refused. Escape cancels either tool. Cropping a picture forgets the assignments and axis values given on it, since its plots and traces are numbered again.

An axis whose labels did not read waits. Click a tick label on the picture, read or not, and type what it says in the box beside it: a y value as a number in the axis's units (-10), an x value with its unit (2.4 GHz). A bare x number takes the unit the other x labels were read in, and is refused when none was read. Two labels on one axis calibrate it; a third replaces the older of the two. This also corrects an axis calibrated from misread labels: the values you give replace what was read. Each value is written into the plot's x axis or y axis row as a pixel position and its value, 120px=1GHz,900px=5GHz, and the row can be typed directly in that form. A logarithmic axis is read as one from its labels; Log beside the row says so when its labels did not read.

A y axis on the right. Many datasheets draw two quantities against one frequency axis — dB(S21) against the left axis and its phase against the right, or gain and noise figure. The right axis is read from its own labels, and the plot's row gains y2 axis (right), which waits and is typed or clicked exactly as the y row is: a label clicked right of the plot gives a y2 value. A plot with an axis on the right only reads it as its y axis, and says (right).

The right column has three panes, each scrolling on its own; drag the grip between two panes to give one more room. Plots is at the top: choosing a plot lists its traces in Traces below and shows its picture, and choosing a trace on the canvas chooses its plot. Output at the bottom holds the matrix, the output grid and the report.

Traces. Each row is one curve: its colour, its range, its points and its resolution per pixel. Entry and quantity say what it is — S21 in dB, S11 as phase, a chart's complex value, VSWR, return loss and so on — taken from a legend entry in its colour, a label along it, the y axis's title or the plot's title, and the dot says how sure that is. A curve nothing names is Unassigned, and the Touchstone waits for you to say which entry it is.

On a plot with a y axis on both sides, Axis says which one a curve is read against, Left or Right; its tooltip says why. It is taken from a mark in the curve's legend entry, from an axis title or labels printed in the curve's colour, or from the curve's quantity matching one axis title — a phase curve against Phase (deg). When nothing says, the left is assumed and an amber dot beside Axis asks you to check it. On the picture a curve read against the right axis is labelled → R. Choosing an axis is undone with Undo like any assignment, and in Simple view it moves every piece of the curve.

Ports, in the Traces heading, is how many ports the file has, 1 to 4. Left empty it is the highest port any trace names, at least 2, shown faintly in the box. The entry lists offer S11 up to that many ports. Lowering it clears no assignment: a trace already given an entry beyond it shows S41 · beyond 2 ports and the Touchstone waits until the entry is changed or the ports raised. Undo beside it (Cmd/Ctrl+Z) puts back the assignments and the ports as they were before the last change, one step at a time.

A curve is often read in pieces — it breaks where it crosses a grid line, a marker or another curve. Simple, at the right of the Traces heading, lists one row per curve colour, Red · 4 pieces · 2.3–2.6 GHz: choosing its entry and quantity assigns every piece. Two curves of one colour over the same frequencies stay two rows. A row too short to be a curve (a legend sample) is listed last, dimmed. Detailed lists every piece on its own row; an assignment made in either view stays in the other, and a Simple row whose pieces differ shows mixed. Pieces given the same entry and quantity are joined into one curve before the Touchstone is made, and the report names them: joined S21 db: T37 + T39 + T42 + T50.

Fit. A curve read off a picture steps by one pixel's worth of value and carries the wobble of the pen's edge. Fit under a curve's row smooths it: Linear, Exponential, Polynomial (with its degree, 1 to 8, 3 when left empty) or Spline. The fitted curve replaces what was read in the Touchstone. It is fitted in the units the curve is drawn in — dB, degrees — against frequency in the axis's unit, or its logarithm on a logarithmic axis. A Spline follows the curve as closely as one pixel unless you type how far it may stand from it, in the curve's own units (0.05). Pieces joined into one curve share one fit, so in Simple view the fit is set on the colour row. A Smith or polar curve is not fitted.

Under a fitted row is its equation with the unit of its variable, -0.51 + 0.012*f - 0.00041*f^2 · f in GHz (a spline says spline · 412 knots), and how far it stands from what was read, RMS 0.012 dB (0.3 px). Under about half a pixel it follows the drawing; several pixels means another model fits better. Copy puts the equation on the clipboard, in the form a variable or a parameter takes. On the picture the fit is drawn thin and dashed over its curve, so a fit that misses shows at once. An exponential that cannot describe the curve says so and leaves the curve as read. Choosing a fit is undone with Undo.

Reading a value off a curve. Hold the pointer near a traced curve and a crosshair snaps to it, with what the curve reads there beside it: T3 · 2.437 GHz · −12.31 dB, or on a Smith chart T3 · Γ 0.41 ∠ −63° · 2.40 GHz (no frequency until the trace has one). The readout follows the curve itself, between the points read, and gives no more digits than one pixel of the picture resolves. With a fit set it adds the fit's value, fit −12.29 dB.

Points. The Point tool, beside the box tools, pins a point where you click on a curve; it stays on until Escape or the button is clicked again, and a click away from every curve is refused. Each point is a numbered dot on the picture and a row under Points, P1 · T3 · 2.437 GHz · −12.31 dB. ✕ removes one, as does clicking its dot and pressing Delete. To read a curve at an exact frequency, select its row, type the frequency in the box under Points (2.4 GHz, as an axis row takes it) and press Pin or Enter: the value there is pinned, worked out between the curve's two nearest points read. A frequency outside the curve's range is refused. Copy puts every point on the clipboard as tab-separated columns — point, trace, x, x unit, y, y unit — ready to paste into a spreadsheet. Points belong to the dialog: reading the picture again keeps them on their curves, and nothing writes them to a file.

Frequency along a Smith or polar trace. A chart has no frequency axis. It comes from markers on the trace whose readouts show their frequency, or from a rectangular plot of the same entry in the session, matched point by point; with neither, the trace's Start and Stop are highlighted and wait for you. Between known frequencies they are spread along the curve by the angle it sweeps round the chart's centre — exact for a plain line, an approximation for anything that resonates — so a chart with three or four markers reads far better than one with none. Frequency runs clockwise round the chart unless the markers say otherwise. Smith Z0 is the chart's own Z0 = when one is printed, else 50 Ω, said in the report.

Matrix. A Touchstone needs every entry, and a datasheet rarely plots them all. The grid shows each entry of the file and where it comes from — plot 2, = S21, zero, min phase — and each is a menu:

Minimum phase works the phase out from the magnitude curve. It is a fair estimate for a filter or a matching network and a poor one for anything with a delay in it, and it is least sure near the ends of the plotted band: the report gives the middle 80 % of the band as the part to trust. If the datasheet also plots the phase, add that picture too — a phase read from a picture is always better than one estimated.

Output. The grid is the frequency range every trace covers, at the finest trace's resolution, rounded to a tidy step; type a start, stop and number of points to change it, and log for logarithmic spacing. The Touchstone is written in dB and angle unless you choose magnitude and angle or real and imaginary. Create… asks where to save it; its header records each picture's name, how every trace was read and every fill and phase rule. With Keep the source pictures beside the file, each picture is copied beside it as <name>.source.png. The new file opens in the Data Display, and its passivity, reciprocity and causality check goes to Messages.

CSV writes the curves exactly as read, in each plot's own units, and never waits: any plot digitises, S-parameters or not — a temperature against time comes out just as well. Its y_axis column says which y axis each rectangular curve was read against, and its last column, y_fit, holds a fitted curve's fit at each point beside the value read. With CSV chosen, Ports, the matrix and the output grid are not shown: they are the Touchstone's.

Digitise by hand

With Read: Manual the right column is six steps, each enabled once the one before it is done.

1. Plot area. Drag a box across the plot's axes. Its left and right edges are x low and x high, its bottom and top y low and y high. Drag a corner or an edge handle to resize it. For a scan that is not square, drag the round handle above the box or type its Angle. When circuitRF happens to find the plot's frame, the box starts there, dashed: move its edges onto the axes. Draw box draws it again.
2. Scales. One y scale or Two y scales (a second on the right). Type x low and x high — a frequency with its unit (1 GHz), or a plain number for an x that is not frequency — and y low and y high, with a unit after the number if you like (-40 dB); that unit is the CSV's y_unit. Log marks a logarithmic axis. Each value is written beside its edge of the box, so a wrong corner shows at once. Text in the picture is never read: the scale you type is the scale.
3. Curves. Add curve for each curve to take. Each row has a name (the CSV's trace column), a colour, Axis (with two scales) and Is: Data — plain x and y — or an entry and quantity such as S21 dB. The row's copy button adds a curve on the same points, for two curves that run together (S12 on S21): move its points where the two part.
4. Points. With a curve selected, double-click on the picture to add a point, drag a point to move it, and right-click it, or click it and press Delete, to remove it. Escape stops placing; Place points starts again. While you place or drag a point a 4× magnifier follows the pointer, and the readout beside it gives the value under the pointer. Find curve, then a click on the curve, follows it by its colour inside the box and places its points for you — one every 20 pixels and one at every peak and dip — to edit as you would your own. A click on blank paper finds nothing, and says so. Undo (Cmd/Ctrl+Z) puts the box, the curves and the points back one change at a time.
5. Sample. The points are joined and sampled at x values: from the curve's first point to its last, 101 of them, unless you type a start, stop and number of points, or a step (50 MHz). Linear joins the points with straight lines (in log coordinates on a log axis); Smooth with a curve through every point that never overshoots between two of them. Nothing is sampled beyond a curve's first and last points. The sampled curve is drawn dashed in its colour over the picture.
6. Create. Create… writes the CSV, one row per sampled point. When a curve Is an S-parameter quantity on a frequency axis, Also save as Touchstone… writes a Touchstone of those curves as well, on the same x values.

The same reading runs from the command line, circuitrf recognize plot.png --manual-box … --x-range … --y-range … --curve … --csv out.csv (the command-line reference lists every flag).

Layers and presets

A picture's colours are its only layer information. Layers lists each colour with its share of the picture and what it is read as: one of your technology's layers, Drill, Board outline, Silkscreen, Background or Ignore. The first reading maps the colour touching most of the picture's border to the background, round holes to Drill, and the rest to your copper layers by area, top first; a colour that is two translucent layers drawn over each other is read as both. A colour drawn as lettering goes to Ignore: silkscreen is read only when you set its row to Silkscreen yourself. Hover over a row to see its colour on the picture.

A photograph is read differently, since it shows one side of a real board and its largest colour is usually the laminate or the mask. Its colours are read with the lamp's tint taken out, the copper-coloured ones go to your top copper layer, and the rest — the board, the bench — to Ignore. Change a row if a colour is read wrongly.

Change a row and it is marked with a dot: every later reading keeps it.

Pictures from one source use the same colours every time, so a map can be kept. Save Preset… keeps the map and the options under a name; Preset applies one. When a picture's colours match a saved preset, the status line offers it — Colours match preset "viewer dark" — and applies it only when you click Apply. Presets belong to you, not to a workspace.

Technology

A picture says nothing about the stackup. The traced layout and the recognised schematic use the technology chosen here — the workspace's default to begin with — and the report says once that the stackup was not read from the picture. Changing it reads the picture again.

Any technology can be chosen, for its layers: the workspace's own, then circuitRF's, marked (circuitRF). One of circuitRF's is installed into the workspace's tech folder when Create writes a cell with it, as a new workspace's technology is; a file already there is never overwritten.

Advanced

Collapsed until you open it, and it stays as you left it. Each change reads the picture again.

The picture under the result

The written layout and schematic carry the source picture behind them, faded to 35 %, placed so each shape lies on what it was read from. A schematic picture drawn Tidy has none, since its parts are not where the picture has them; one drawn into the schematic it is placed in needs none, since the picture is already there. Checking the result is then comparing in place. The picture is copied into the cell's folder; it is an ordinary bitmap, which you can move, resize, fade or delete like any other — select it and Properties shows its file, position, size and opacity — and it is never exported, checked or meshed. Clear Keep the source picture under the result to leave it out.

With the picture kept, Create writes a layout even when no copper is traced — set every colour to Ignore and it holds the picture alone, at the scale you set, to draw over by hand in the layout editor.

What reads well, and what does not

Getting a good picture

• Crop to the circuit. A title block, a page of notes or a second circuit beside the first is read too.
• PNG rather than JPEG, and the original rather than a copy of a copy. A JPEG's smudged edges are read, less exactly.
• Big enough. Lines 2 pixels wide or more, and text at least 14 pixels tall to its capitals. Zoom in before taking a screenshot rather than scaling the picture up afterwards: scaling adds no detail. A circuitRF schematic reads well exported at 3× and above.
• One colour per layer on a layout, distinct from the background and from each other; drill holes drawn as filled circles. A drawing that shows its holes in the background's colour is read too: a round hole in the copper up to 1 mm across becomes a via, with the copper whole under it, and a wider one — a mounting hole — stays an opening. A hole drawn with a dark rim is one hole, drilled on the rim's centre line, and the rim is not copper. Drills and pads are rounded to a whole mil, or to 0.05 mm, in the technology's unit. A via whose picture draws no pad takes the technology's via pad, or twice its drill if that is smaller.
• A line drawing — copper drawn as outlines — reads no copper from its lines, and says so: which of its closed outlines are copper, and not a hole, a part's body or a letter, is not in the picture. Crop to the board, then click inside each pour, pad and trace with Paint Fill; the round holes inside each become vias. Draw the Outline so the scale row gives the board's size to check. A pasted picture needs no file for this. Painting needs no scale, and a line painted in is one Line impedance… can set the scale on: its width is measured between the drawn line's centres.
• Something of known size on a layout picture — a scale bar, a dimension, a part whose size you know, a line of known impedance — so the scale can be set. A picture of ordinary chip parts often sets its own.
• If you have the original file — the PDF, the SVG, the board's Gerber — it is the better source. A board's files go to Create Schematic from Artwork; a vector drawing, exported as a PNG at a high resolution, reads better than a screenshot of it.

When nothing can be read

A picture with no drawing in it is shown faded, with one sentence at its foot saying what was seen — No drawing found: a photograph-like picture with no straight line work, a blank picture, a page of text — and two links, Read it as a schematic and Read it as a layout, for a picture you know better than that. A picture of line work that holds no circuit — a block diagram, a drawing — is read as a schematic and refused the same way: No circuit: most of the lines in this picture curve or slant, with the one link that reads it as a layout. A reading that finds nothing to make — no copper at the mapped colours with Keep the source picture under the result cleared, no technology — is shown the same way, with the control that answers it highlighted. Create stays disabled, and its tooltip is that sentence.

Troubleshooting by report line

What it does not read

• Hand-drawn schematics are not a goal; a neat one may read.
• Photographs of printed pages, of a board's underside aligned to its top, and X-ray pictures. A photograph of a board's top reads its lines and pads only once set square on its corners.
• Script type, and type that is very small or slanted at a small size. A value printed in such type becomes a variable or, rarely, reads wrong: check values against the picture.
• Some datasheet drawings — see Reading a circuit from a datasheet or a textbook.
• A value printed far from its part, other than on the line under its designator, may be given to the part nearest it.
• Dense layouts where a trace is under about four pixels wide: widths are read, but loosely.
• The stackup. A picture never states one; the technology you choose supplies it.
• Vector files — PDF, SVG, EMF. Export the page as a PNG.
• Vias drawn as two thin rings, as circuitRF's layout view draws them, below about 10 µm to a pixel; draw drill holes as filled circles.
• A part on a short stub to a via, with no solder mask in the picture, may not be found; a part in a line is.