10 Low Investment Business Ideas for Beginners in 2026 (Start Under $500)

May 1, 2026•14 min read•By Maker's Workbench

MakerSMTCoreXYMechanical

Build a DIY Pick-and-Place Machine for Solo PCB Assembly

A real, working machine on your bench that places 0402 resistors faster than any human and never asks for coffee. Cost: $700. Time: a long weekend plus a week of fiddling.

Why build instead of buy

Commercial desktop pick-and-places start at $4,500 (LumenPnP, with feeders) and the entry-level pro models from China land near $9,000. If you can place 0805 and SOIC reliably, that's all most maker-tier products need — and a CoreXY frame with vacuum tip and a USB camera does that beautifully for under $700.

Honest expectation: a DIY machine will hit 1,200–1,800 components per hour with one nozzle and one camera. A commercial 4-head machine does 8,000+. Plan accordingly. For 50–500 units of a small board, this is the sweet spot.

Bill of materials

Subsystem	What	Cost
Frame	2020 aluminium, 600 × 600 mm	$70
Motion	NEMA 17, GT2 belts, linear rails MGN12	$150
Z + nozzle	Hollow stepper + vacuum nozzle set	$60
Vacuum	12 V diaphragm pump + solenoid	$45
Controller	Smoothieboard / Octopus Pro	$70
Vision	Two USB endoscope cams	$30
Feeders	3D-printed strip feeders ×8	$25 PLA
Power, wiring, fasteners	—	$80
Software	OpenPnP (free)	$0

Mechanical: the CoreXY trick

CoreXY decouples X and Y motion onto two stationary motors with crossed belts. The moving mass is just the nozzle head, not a full gantry. That keeps acceleration high (3 m/s²+) and ringing low — both critical for placing 0402 parts without flicking them off the tape.

Two belts, two stationary motors. X = (A+B)/2, Y = (A−B)/2. Beautiful.

Vision: down-cam + up-cam

Two USB endoscope cameras at ~$15 each. The down-cam (mounted next to the nozzle) reads fiducials on the PCB and corrects X/Y/θ. The up-cam (fixed to the frame, looking up) measures the part on the nozzle and rotates it to alignment. OpenPnP does the math — you supply lighting (a $4 white LED ring) and good focus.

Feeders: where most builds fall apart

The motion is the easy part. Reliable feeders are not. The lowest-effort path: print 8-mm strip feeders that the head physically advances by pressing a lever between placements. It costs you ~80 ms per pick — fine. Only build motorised electric feeders when your throughput justifies it (over 2,500 cph or 12+ part types).

OpenPnP: the brain

Open the package, edit machine.xml, and define your motors, nozzle tips, and cameras. OpenPnP's job-runner reads your PCB centroid file, sequences placements to minimise travel, runs vision alignment per pick, and logs every placement.

# job_recipe.yaml — minimal
board: rev3.kicad_pcb
fiducials: [F1, F2]
parts:
  R_0402_10k: { feeder: F1, qty: 24 }
  C_0402_100n: { feeder: F2, qty: 30 }
  ATtiny_QFN: { feeder: F8, qty: 1, vision: precise }
optimise: nearest-neighbour
report: csv

Where it earns its keep

Three margins:

• Production runs of 50–300 boards at home instead of ordering JLCPCB assembly (you save $1.20–$2.50/board, including shipping)
• Quick turn revisions in 2 hours instead of 5 days
• Service offering: charge $0.04/placement plus $1/board for friends and small startups in your local maker scene — it pays for itself in 4 jobs

What goes wrong (so you can plan for it)

Tip pickup failures (vacuum leak, dirty filter), parts knocked over by accel ringing, fiducial mis-detect under glossy soldermask, feeders that feed two parts at once. Each has a five-minute fix once you've seen it. Budget 8 hours of "first job hell" before your machine starts paying you.

Start now: assemble just the X/Y axes this weekend, command them via Pronterface, and watch your nozzle move. Everything else is iteration around that core motion.