I have you, Steve to thank (blame) for this latest obsession. OK, you and the world-wide resurgence of interest in Analog Synthesizers, combined with my love of analog designs and music.  In the mid 70's an old friend Jeff, a very smart analog engineer, designed and built an analog synthesizer from scratch. It was a modular, and had most of the features you would expect. I played with it a bit, not much. Fast forward 40 years, and the world has re-discovered these things. Moogs and Arps sell for many times over their original prices.

I put this off for a while because I had little to offer the synth community. In fact, most of the circuits were designed 40-50 years ago and have been refined ever since. One big problem with old synths and their designs is that the opamps, transistors and other components used to build the original machines have long gone obsolete. Originals and replacements are old stock only and sell for top dollar. Meanwhile, there are few surface mount designs. Surface mount offers smaller, more modern and thus available parts, and machine assembly as well as hand assembly. As a pro-fessional 'lectical engineer, I almost never design with thru-hole technology.  Even my test boards and one-offs use SMT. The small size offers lower parisitics and thus closer to theoretical performance. Not so much an issue with audio frequencies, but it doesn't hurt. I find it easier to mount and solder a part on one side instead of bending the leads, mounting it through the board, bending the leads again to hold the component in place, soldering it through the board, flipping the board, soldering the leads on the rear, then trimming the leads. It's not so bad, but how 70's.
 
Here is one example: do a Digikey search for 'matched NPN' then filter by SMT and by thru-hole. There are two old and expensive part numbers in thru-hole and a couple dozen in SMT including the precision SSM and MAT series from Analog Devices. Some SMT parts that are matched to better than 2mV are less than $.50. No more hand matching and gluing together transistors. Thank you, modern technology.

There are many modular synth systems, starting with the Moog Modular and others from the '60s. The current rage is Eurorack 3U. An amazing assortment of Eurorack modules are available in kit, board or completed form. The Eurorack platform is very simple, a 16 pin ribbon cable backplane provides power and a couple of bussed control signals. Mechanically, each module needs only a flat front panel and a few screws. The card cages use standard hardware, and are strong, flexible, low-cost and available.

The concept of modular solutions has always appealed to me. When I was at WPI in the early 70s, I learned to simulate physical systems on Analog Computers. Yes, I am that old. In the 80's at Datacube I was one of the inventors of generations of MaxVideo modular image processing systems. At Analogic and Teradyne, we developed modular instrumentation for PXI, VXI and several proprietary platforms. At home I worked on modular solutions for microprocessor development and for low cost instrumentation solutions, but these never got off the ground.

One appeal of Eurorack is its utter simplicity. Each module is sheet metal,  a circuit board or two, a ribbon cable, some knobs and jacks, and the creativity you bring. I love it.


Here is my first pass synth system on the electronic work bench, in all its wiry glory.  From Left to Right:  MIDI2CV, Sequencer prototype, Dual VCO, three ADSRs, VC-LPF, and a passive Mixer. It is beginning to look like a synth, but still has a science experiment look.

VCO

I wanted to get started with some existing technology and to gain some experience with the basic mechanical components, pots, jacks, switches, etc, as well as with building and packaging. I looked for VCO, VCA, VCF, and ADSR designs that I could lay out in SMT, and that would give me full function quality electronics. The VCO is one challenge. Since it is the heart of any synth, it needs to be flexible. Accurate voltage control, manual pitch, FM, and voltage control over parameters such as pulse width are all useful functions found in most full-featured VCOs. Waveforms should include all the basics: Sine, Triangle, Ramp and Pulse. It needs to be stable with temperature and time so that continuous tuning is not required. Most VCOs use an exponential voltage to current circuit that was originally published by National Semi in the '60s. The 'standard' for voltage control is 1 volt per octave (1V/O) and it needs to operate over the audio range of about 20Hz to 20KHz, or about 10 octaves. Stated as 1000:1 it sounds easier. Most designs use the exponential current source to charge a capacitor, typically until it reaches a threshold, then the capacitor is discharged. Charge and discharge the cap at the same current but opposite polarity, and you get a triangle wave. Discharge it quickly and you get a ramp, AKA sawtooth or saw.

A ramp can be converted to a triangle using full-wave-rectification, also known as the absolute value function. A triangle can be converted to a sine with a non-linear circuit, and a triangle or ramp can be converted to to a square wave or varying pulse width via a comparator.

Turns out a triangle can be converted to a ramp using an inverting-non-inverting circuit. So the question is, which waveform to begin with, a ramp or a triangle? There are numerous examples of both. Converting a ramp to a triangle is harder than it looks. The fast rise-time of the ramp causes a fast glitch on one peak of the triangle, which causes unwanted harmonics. This glitch will show up on sine waves derived from the triangle. Just what you don't want on triangles and sines:  harmonics. I have seen circuits with two trimpots to try to null out the glitch but that seems like a hack. So I decided to start with a nice mellow triangle, and then use a fast switch to add the fast transients that a ramp needs.  I came up with the idea of using a current mirror to invert the current, and CMOS switches or transistors to route the currents.

It also turns out that the symmetry of a triangle works better than a ramp when operating at high frequencies. A ramp's capacitor discharge takes a finite amount of time, and that time is added to the desired ramp period, causing an error in frequency that gets worse at higher frequencies. To fix this, most VCOs have a high frequency adjustment which is not a simple adjustment to make. With a triangle, the time to switch direction is due to a comparator and a flip-flop, generally faster than a transistor discharging a cap. I haven't implemented a high frequency trim yet, and the VCOs work well without it. Also if my 10KHz is a bit off, who will know? Most dogs don't have perfect pitch. That was the theory, and the Spice simulation agreed. When I built an dtested four of these VCOs, I found that the high frequency linearity was excellent up to 16KHz. If anything, the oscillator was a bit sharp (too high frequency) at 16KHz, or about C6, without any compensation trim. Good enough. I suspect this is an advantage to the triangle VCO design vs. a ramp design.

For the control logic, I pictured something akin to a 555 timer. A 555 compares a voltage (usually on a capacitor) to 1/3 and 2/3 of the power supply voltage, and sets or resets a flip-flop. It is basically a triangle and square wave generator. I did a google search for triangle VCO and came up with Ian Fritz page. His schematic for the triangle VCO looks a lot like my pencil block diagram. http://home.comcast.net/~ijfritz/sy_over.htm He uses a "Wilson current mirror" and CMOS switches to get the current inversion accuracy and symmetry, and used a couple of comparators and CMOS switches to replace the flip-flop. His design is nice but uses kind-of a lot of parts. Bergfoton also has a transistor based Triangle VCO here. Again, kind of a lot of parts and trims.

Then I thought of using an OTA, the LM13700 as the switchable polarity current switch. These operate over 5 or more decades of current, and I only need 3-4. I googled "triangle VCO with LM13700" and found Thomas Henry's excellent "555-VCO" on Muffwiggler. This is just what I had in mind. The core oscillator is half of an LM13700, a buffer amp for the triangle, and a 555 timer. No symmetry trimpots,  and minimum component count. Then he cleverly uses the Reset pin of the 555 as a VCO 'SYNC' input, and the TRIG output to switch the gain of the ramp generator. Very clever design using all the features of the little 555. To convert the triangle to a ramp, he uses the 555 trigger output to switch the polarity of an op-amp. I thought this was pretty genius, however a few weeks later I was perusing some old 1960's EML synthesizer schematics to see how their VCOs worked, and there was the same triangle to ramp circuit using the same inverting/non-inverting circuit.  Many good ideas have been around for a long time.

Thomas published his design and offered PC board builds including a Eurorack front panel for short money. Unfortunately I missed his buying window. Besides I'd like to lay out and build the boards myself. I looked in my basement and found a couple of ancient LM13600's, TL084's and CMOS 555's. In a few hours I had the basic VCO working on a solderless breadboard. I tried a couple of triangle to sine circuits and they worked OK. Henry uses the other half of the LM13700 to make a sine wave converter. One problem with sine wave converters from triangles is that it is hard to remove the pointy top and bottom. If you run the amp at higher gain, you get a flat-top sine which adds odd harmonics. What he does is to subtract some triangle from the sine to null out the peaks. This works pretty nice.

I changed his design a bit:
    Converted it to surface mount. Mine uses one board with 2 channels instead of two larger boards (one front panel board and one electronics board) for just one channel.
    Smaller front panel so I could get more in a rack. His is 16HP, mine is 9HP per VCO.
    Used a low-cost SMT matched-pair PNP. No matching or gluing.
    Added a ground plane (easy with surface mount)
    Used low-cost 16mm pots
    Used a 3 position switch for FM input: Linear, Exponential, or none
    Eliminated trimpots wherever possible. Only the one Volts/Octave trim is really needed.

I laid out the board with ExpressPCB on a Mini board. This is 3.8" x 2.5", a reasonable size for Eurorack systems, and only $75 for 3 boards. I plan to build a full synth this way on about 3 PC boards.

Not one to waste PC board area, I put as much electronics as possible on one PC board, and then hand wired the front panels. But when I built up a couple of dual VCOs with 10 pots (only 5 are PC mounted) 20 jacks and 2 switches, I found that the front panel wiring was a *lot* more work than soldering PC boards. The boards were straight SMT assembly, but the front panel required many hand built connectors and cables, and lots of hand wiring to the numerous pots and jacks. I decided that this was not a lot of fun. Wiring the first VCO front panel was pretty exciting. After that it was drudgery repeating the work 3x.

I had begun laying out an ADSR and a dual VCA, and decided that instead of one board and lots of hand wiring, to use two parallel boards for each module. One for the front panel controls and one for the circuit. This was how the Midi2CV was built, and it was fun and easy to build. I changed from 17mm pots to 9mm, and from hand-wired jacks to the Earthervar PC mounted types. These parts work pretty well and are low cost, but their PCB mounting heights are not the same. So I soldering the tallest parts first, the jacks, and then mount the pots to the panel so their pins aren't fully inserted in the PCB. (see photo).

While debugging it I came up with a few issues. I decided to use two 1V/Octave inputs to allow the sequencer and the keyboard to drive it at the same time. To do this, the two channels should match really well. So I matched the two 100K input resistors to about .1%. Next time I'll just buy .1% resistors since they are cheap enough. The matched PNP transistor pair I selected, PN5201, comes in a very tiny package, smaller than a sot23-6. I was able to just barely fit the pins to the pads. Then I found a nice low cost matched pair of 2N3906 PNPs, in a SOT23 package.  I'll use these or their 2N3904 NPN brothers for future designs that need matching.

When I looked at the sawtooth waves, there was a good-sized step in the  middle of the ramp. Worse, it varied from board to board. My prototype had a tiny step, so what had changed?  I had used Intersil ICL7555 CMOS timers on the prototype and TI's TLC555 on the PC board.  The 1/3 - 2/3 voltages on the TLC555 parts are not as accurate as the 7555s, and so caused DC errors in the triangle wave, that then caused glitches in the ramp as well as offset errors on all the other waves. A quick order to Digikey for ICL7555s, and the problem was fixed. There are still visible glitches, but much smaller. To really reduce the glitches,  the original design uses a trimpot which I wanted to remove. Below, I discuss other '7555 problems found when I built the first sequencer.

I wanted an LFO and realized that it should probably have the same basic waveforms as a VCO. More even. Ramps of both polarity sound the same, but  opposite polarity ramp VCOs definitely sound different. I found that the basic VCO works down below 1Hz, low enough for now. So the VCO works as an LFO without increasing the core capacitor. Cool!  To operate it as a really low frequency VCO I probably should add a switch or change the cap. With the second VCO, I can experiment with using one as an FM source for the second from low frequency modulation (tremolo) to high frequency (FM synthesis), also mixing a fundamental and one harmonic. Fun with music.


Dual VCO front view

Dual VCO rear view, with lots of hand wired components.

Dual VCO PC board

One channel of Dual VCO Schematic

MIDI to CV

Output Precision

The voltage outputs of the MIDI2CV are quite accurate and operate from 0 V to 8.00V,  9 octave range. However, there is an output protection resistor of 220 ohms on each CV output. This causes a slight error of -.2% in voltage when a 100K VCO load is applied. The bad news is that when you drive multiple loads, each load causes an additional -.2% error. At high frequencies, this error is a larger pitch error. Say 1% of 8V is -80mV. at 1V per octave, an error of a full semitone. Since I use my MIDI2CV regularly to calibrate and test VCOs, and to drive multiple VCOs and filters, and possibly other stuff as well, I plan to address this error.  An external precision buffered multiple would address this, but even it will have some input load. There are a couple of ways to address this. The resistor is there to limit the current output in the case of a short circuit. Just plugging in a cable to a jack causes a momentary short to ground as the tip of the plug slides past the ground ring of the jack. If you were to mistakenly partially plug one in, a short circuit could be present for a long time. The CA3140 op-amps specify a maximum short circuit output current as about 50mA so the power dissipated in the op-amp during a direct short to ground is about 12V x .05A = .6W, quite hot for a plastic 8 pin DIP package. The 220 ohm resistor reduces this power by about 1/2, a safer value. Like many op-amps, the CA3140 is specified to drive a short to ground or to either supply voltage continuously without damage. So I may just replace the 220 ohm resistor with a short circuit. If I do this I'll have a few spare CA3140s around in case they get damaged.

A more elegant solution is to take the feedback of the circuit from the output of the 220 ohm resistor instead of the output of the op-amp. This way, voltage drop across the resistor is compensated for. This requires cutting the traces to the wipers of the gain trimmers and adding 4 wires.  Then because the 220 ohm R and any cable capacitance forms an additional high frequency pole to the feedback, an additional compensating capacitor may be needed from the op-amp output to the - input. Cable has about 10-20pF per foot, and maybe 10 feet of cable to multiple loads,  so about 200pF of capacitance worst case. I should be able to apply, say a 470pF cap to the output and test the step response.  Then add compensation if there is excess overshoot, ringing, or oscillation.

As a shortcut you could apply this fix only on the main output,  CVA and then make sure to use this one for heavily loaded or precision patches.


MIDI2CV, built from a kit from HexInverter

Sequencer

My keyboard skills are limited to say the least, and I wanted a way to play simple sequences automatically, without having to hit keys. So I investigated sequencer circuits, initially just as a test generator. I like the old SimpleSeq from Hexinverter. Their clever circuit uses a 3 way toggle switch per note. The 3-way switch positions are ON, OFF and Loop, allowing the sequence to be set from 1 to 8 notes and to skip any notes. The clever part is that diodes connected to the pots and switches control the analog output, gate, and reset. I cut a front panel and wired it up. Again, lots of hand wiring, I built the first one using eight old 10 turn pots I had lying around. This was a mistake since the 10 turn pots don't have an indication of the note settings. I rebuilt it with smaller 16mm single- turn pots with indicator knobs, which work better, and are smaller and cheaper. Again, lots of hand wiring. This really needs a PC board.  I built a second panel with smaller one-turn pots. It is a work in process.

For its gate output, it occurs to me that having a variable width output would be useful. Most sequencers use a variable clock frequency and a fixed  50% duty cycle. It's a bit tricky to build an oscillator with independent frequency and duty cycle outputs. I picture a triangle oscillator and a variable threshold to set the duty cycle. Someday...

Here is the current sequencer schematic, a work in progress. One interesting bug I found. At first I used an ICM7555 for the clock. The circuit regularly skipped notes in the sequence. It turns out that the 7555 output occasionally had glitches on one edge transition, maybe when operating at slower clock speeds. I tried lots of grounding and bypassing tricks, but couldn't stop it. I substituted it with a TLC555 from TI, and no glitches. So this circuit works better with the TLC555 and the  VCO  is more precise with the 7555. Well, Professor Eaton at WPI in 1976 warned me against using '555 timers, and I generally heed this advice. However they are particularly appealing for quick and-dirty synth circuits. Let the designer beware...

VCA

VCF

ADSR

I chose the circuit from Music from Outer Space (MFOS) by Ray Wilson. Their boards are not Eurorack, and I wanted a few of these, so I laid out a PC module with two parallel boards. With an ExpressPCB mini board (3.8" x 2.5") I built a front panel board and a circuitry board, each about 1.2 x 3.8", and then cut the two boards apart lengthwise. This worked out pretty well.

But when I built them up, They would work sometimes, and not others. I checked out the circuit, and discovered that the TL074 op-amp was getting hot when it stopped working. This corresponded to an increase in the -12V power supply current to the system. Turned out that the CMOS 4066B switch doesn't like any negative voltage on its input pins, and would latch-up. This only occurred when the -12V supply came up at the same time or before the +12V. My system, like many Euroracks, uses similar + and - 12V supplies, and since the negative supply is generally less loaded than the +, it comes up a millisecond or so before the plus. You could make it occur or not occur by plugging in the 10 pin power connector to the ADSR tilted towards the + or the - end first. I sent Ray Wilson a note with this problem, and he said that hundreds of his ADSRs have been built without any reports of this bug. I suspect it may be due to the specific 4066B manufacturer parts that I used vs. the ones he uses. I fixed the problem by changing the TL074 to a +12V CMOS single supply part. This required a cut and jump on the opamp's - supply pin. It works fine this way and I have built 3 of these so far.All three failed when I used the original TL074.  In general, circuits that use multiple sourced parts should not depend on a specific manufacturer's part.

Here are some shots of the ADSR board. Note the stacked board construction, the use of Earthernvar 3.5mm jacks, and 9mm Pots. No hand wiring, Yay!


The board is an ExpressPCB mini, sliced down the middle to make the two boards. PCBs for 3 ADSRs for $80, not bad. 

Building Front Panels

Current Status, October 2015

• One Rack, Homemade chassis, Vector rails and Synthrotek Power Board
• Midi2CV , Hexinverter kit
• Dual VCO, modified Thomas Henry design, my PCB
• Sequencer, half-built, based on SimpleSeq from HexInverter
  
• VCA, still on a solderless breadboard, based on MFOS VCA
• ADSR, MFOS ADSR, but my PCB layouts
• VC-LPF, Synthrotek
• Passive mixer: Just jacks and 100K resistors
• Power Supply. I'm still using a dual lab supply to power this.
  

• Sequencer: Wire up all the gate logic and design a variable width gate. Maybe bring out all the individual outputs to drive a drum machine.
  
• Dual VCA board, constructed like the ADSR on an ExpressPCB mini board, cut in half
• Noise Source, maybe a random sequencer like Turing Machine
  
• Two more VCO/LFOs: another dual just like the other
• Proper Mixer with amplifiers.
  
• Full function filter: High, Low and Band-pass
  
• Copy of the Moog Ladder filter? I Want to see what the magic is all about.
• Play with advanced Midi2CV features like keyboard split, polysynth...

11/1/15

Prototyping and Design:

Mixer

Case: Rev2

I have pretty much filled up the first 19" case, and foresee filing another with a couple of VCOs, VCAs, a better sequencer, not to mention some cool commercial modules. Three 19" racks should last me a few years. Plan A was to buy or build 3U rack sides, but I don't really want to mount it in a 19" rack. Instead wooden sides for now, maybe a handle on top,  designed for small size and portability. I don't have any other rack stuff, so wood it is. Since this will regularly sit behind

That means the rack sides should be simple metal strips to hold the Vector rails. These will have a couple of holes so they can be screwed into the wooden sides. A few caveats though: First, you need to remove most the modules to access the screws. I don't foresee removing the racks many times, but I may eat those words.  Next, Power distribution boards need separate mounting. To ease the pain, I plan to use a single board (Synthrotek, but apparently obsolete??) with 17 16 pin connectors, located between the two 3U racks so the cables can reach modules in 2 racks. Finally, I discovered that the screws that mount the rack to the wood need to be flat heads and countersunk. Pan head screws interfered with some modules.

Here it is new rack. The power distribution board is obscured behind the center rails. Note the new Synthrotek Mixer.  I built up the second Dual VCO this weekend.  The blank panel will be the third VCO,  someday. Even with the Dual VCO PC board 99% built, it took about 7 hours to machine the panel and hand wire the 18 jacks,  2 switches, and 5 pots. Still have only the one proto-board VCA, off screen. That's next.

Update 11/15/15

Dual VCA

Finally the VCA boards arrived and I built up the first one. The linear gain control works fine, but I got no output in Exponential mode. After a bit of debug, I found that I had bought the wrong matched PNPs. There are two versions of the DMMT3606 and I bought the wrong one. Worse, the only matched 2N3906 in a SOT23 case has the bases in common, not OK for an exponential source, and the correct part only comes in a SOT363, which is the tiny package of the PN5201s. I need to re-lay out the VCA and VCO to use this tiny package. I hate to use such a fine-pitch part. Sorry.   

The initial output waveform of the VCA was a bit distorted. When 10V p-p (+/- 5V) triangle input was applied, there is too much curve on the slopes of the triangle (PHOTO!!). This is from overdriving the low-voltage inputs of the LM13700 with a 25K/500 ohm 50:1 divider, so +/- 100mV at the inputs. They are only linear to about +/- 50mV or less.  A 100:1 divider worked well.

I am getting better and quicker at building panels. With careful setup, the table saw is able to hold about 10 mils accuracy on three panels. Then I was able to mark one panel, drill all the holes to 1/8", then clamp all 3 panels together and using the first one as a template, drill the other two to 1/8". Then it is a simple matter to enlarge all the holes to 1/4", the jack and switch size, then enlarge the Pot holes to 5/16". All 3 panels came out perfect, and in record time.

Here is the Dual VCA schematic. On my first board I installed no trimpots and it works well, but I have not tested the important parameter of feed-through. When the audio input is 0, and the control voltage changes, there should be no output glitch. This is what R13 and R33 are for. I'll let you know.

Synthrotek MST Noise / Sample and Hold

Synthrotek DS-M Drum Machine

Turing Machine Random Sequencer

I am so impressed with the simple Turing Machine sequencer design and the amazing range of sounds it is capable of. Consisting of a simple 16 bit shift register, a noise generator, 2 switches,  a knob and wow! Steve uses one with the optional Voltages and Pulses Expanders as his main sequencer. He uses Pulses to drive drum machines. I need to get one.  The 4 boards and panels are reasonably priced on at $120 on Synthcube. Speaking of which, Synthcube, where have you been all my life! They offer hundreds of DIY kits, boards and panels. Yikes! I am waiting for a couple of the main boards from Thonk for cheap, but have also ordered the full set of 4 PC boards and 3 panels from Synthcube since I really like their panels and want at least the Pulses Expander.

The first board arrived from Thonk. I had ordered parts for it so was able to build the board in a couple of hours. It worked right away. Glad I ordered the other boards and front panels.

Then the other boards and panels arived. The expander and Pulses modules are no-brainers. Very simple and low parts cost. Voltages uses 8 Alpha slide pots with cute LEDs, but these are only $2.50 each from Mouser.

As usual, I am the bug finder!  In looking at the Turing machine schematic. I see that the designer built an XOR gate the hardest way possible: using two transistors and two transmission gates. And I spied the usual bug: An op-amp operating from +/- 12V  feeding CMOS running from +12V. Sure enough, the op-amp gets very warm when outputting - voltage, due to the CMOS protection diode. At least it doesn't latch up and die like the MFOS ADSR did. A simple series diode and a resistor to ground addresses this. People, please don't do this.

It is still not working 100%.  I accidentally ordered the SMT version of the 4015 shift registers. And the ancient ones I had since the early 80's don't like driving the LEDs, so they fail after a few hours. New parts should fix this.

Here is my baby with 2 racks full. Time for the third rack and a second power board.

Power supply

With all the new modules, my power supply is drooping a bit. The problem appears as annoying 120Hz FM on the VCO outputs. I used a +/- 12V adjustable linear board I bought on Ebay. I feed it with a 2x12VAC 1.5A transformer, but 12VAC is not enough to prevent droop at currents of 0.5A. I could use a 15VAC transformer, but then I would be dropping several volts more across the regulators. Ideally it would be  13VAC or 14 VAC per side, not common voltages. Time for a switcher.

I had some old DC-DC boards at work and modified an obsolete one to convert +12V to -12V. For the +12V I use the amazingly small and fully enclosed TDK-Lambda LS25-12, 12V at 2.1A, $20 at Digikey. Works swell.

Synth Kit and DIY Building Tips

• A decent temperature controlled soldering iron with 0.060" chisel tip and 0.020" or 0.025" lead solder
• Nice needle-nose pliers, diagonal cutters and wire stripper
• Nut-drivers for the front panel jack and pot nuts, typically 5/16" and a 10mm deep socket
  
• A bin of plastic drawers to keep parts in
• Illuminated magnifier
  
• SolderWick size 1 and 2
• Good fine tweezers
• Cheap ribbon cable crimp tool
  

The kits I build use a lot of the same parts over and over, so having a stock of the basics is good. You can build this stock up for about $100:

• DIP ICs, a few of each: LM13700, TL072, TL074, LM324
• Transistors: 2N3904, 2N3906
• Resistor 1/4W 5% assortment: 5 or 10 of the common values, check on Ebay and Amazon
  
• Resistor 1/4W 5% in multiples of x1.00, x2.00, x4.99, values from 1.00K to 1.00M. Buy 50 each of the 1.00 values and 20 of the others.  You can always use a 1% in place of a 5%
  
• Ceramic caps, radial, 50V, 0.2" lead spacing, 100p, 1000p, .01u, .1u,
  
• Electrolytic caps 10uf 25V, radial, < 0.3" high
  
• DIP IC sockets, 8/14/16 pins
• Synthrotek or Earthenvar vertical 3.5mm jacks and hex nuts, a bag of 50 will get you started
  
• 9mm knurled shaft pots, Linear taper, 50K, lots of 100K, 1M
• SPDT toggle switches, ON-ON, ON-OFF-ON
• 16mm knurled shaft pots, 100K and 1M for hand-built stuff
• 5, 6 and 10 pin single row female headers, 0.1", gold
  
• Break-apart male headers, 0.1", single and dual row, gold
  
• 10 pin Box headers for Eurorack power
  
• 10 and 16 pin ribbon cable connectors and 10 pin ribbon cable if you  build your own power cables.
  

Then when I build a kit, I go through the BOM and my stock to see what parts are missing and do a Mouser or Digikey order for the missing ones. As I build a board, I print the BOM and mark any components that I am running low on. I generally keep an open shopping cart on Digikey.com. If you go back and touch the cart once a week or so, it seems to stay there. If you're worried about your cart disappearing before you check out, just print out a copy.

Of course you'll need synth cables and adapters:

• 6", 12", 18", 24", 3.5mm mono or stereo patch cables
• 2:1 adapters
• 3.5mm to 1/4" adapters or cables.
  

For hardware and panels:

• M2.5 x 6mm screws for Eurorack
• 4/40  x 7/16" or 11mm standoffs

For building panels:

• 6061 aluminum, 0.062"
  
• 100 tooth soft-metal carbide 10" blade (Amazon)
  
• Table saw for above
• Mill files
• 1/8" Drill bits
• 1/4" Drill bits
• 5/16" Drill bits
• Exacto Knife
• 3 mil Laminator pouches and a low cost Laminator
  
• DAP Contact Cement

To build SMT boards:

• Same ICs, resistors and cap values as above, but 0.050" pitch ICs and 0805 discretes

Surface Mount (SMT) Building Tips