Building a DIY Computer Based on FPGA

This is the story of my project to build my own computer with an FPGA from scratch. Instead of using off-the-shelf boards, I designed the electronics, laid out PCBs in KiCad, and soldered the components myself. The goal was to create a device capable of running Linux with a connected display and keyboard.

What Was Done

The project includes several stages:

• Hardware — designing the PCB with an FPGA and all necessary peripherals, ordering from China, and soldering at home
• SoC (System on Chip) — using existing IP cores (including the VexiiRiscv processor core) with custom components in Verilog and SpinalHDL
• BIOS — written from scratch, including initialization, memtest, RISC-V SBI implementation, command line, and Linux bootloader
• Linux drivers for custom peripherals
• TTY implementation with ANSI escape code handling for ncurses

All code, schematics, and files are available on GitHub.

Background

I had experience with FPGAs from 10 years ago, when I used a Marsohod2 board and tried to create my own processor. The current project started a year and a half ago when I rediscovered that board and decided to build something more ambitious.

My previous soldering experience was limited to microcontrollers in DIP packages. This time I decided to tackle more complex components.

The First Prototype

Planning

Given my lack of experience with BGA packages, I chose something "as complex as possible but still realistically solderable" — the MAX10 series FPGA (10M50SAE144) with 50K logic elements.

Components:

• Memory — four DDR1 chips AS4C64M16D1 (128 MB each) in parallel
• Keyboard — USB connection via two data lines with 15K resistors to GND
• Display — HDMI output, based on examples from Marsohod3
• Storage — SD card (following the fpga4fun.com schematic)
• Other — LEDs, buttons, DAC for audio, Ethernet jack

Decoupling capacitors were added for power stabilization, along with removable jumpers for connecting voltage regulators (5V → 3.3V → 2.5V).

Soldering and First Problems

I bought a hot air station and started testing. The first attempts revealed many difficulties:

Mistakes:

• Too much solder paste caused bridging (short circuits)
• Forgot to connect CC1 and CC2 resistors in the USB-C connector to GND (5.1K), which prevented charging from working
• DC-DC converter U7 was soldered upside down (the chip had an alternative pinout)
• Linear regulator AP2114HA-2.5TRG1 — the letter "A" indicated an alternative pinout, which I initially missed
• Traces were damaged while soldering the HDMI connector
• Wrong FPGA pins were chosen for video output (Quartus required specific pins with DDR IO support)

Successes:

• The programmer (FTDI) worked on the second attempt
• The FPGA was detected via JTAG after a day of cleaning pins
• SD card reads successfully
• LEDs blink

Failures:

• HDMI display shows "No signal"
• DDR1 memory doesn't work
• Audio was not tested

The first board was lost after six months of use — the memory failed, likely due to improperly functioning power supply capacitors.

Second Board Version

Key improvements:

• One memory chip instead of four (higher probability of success)
• FPGA and memory on opposite sides of the board for shorter traces
• Built-in programmer on a single board (verified working)
• Added two Si5351A frequency generators (configurable via I2C), since a single PLL in the 10M50SAE144 wasn't enough
• Redesigned audio output (stereo 12-bit, 44.1 kHz)
• Reoptimized HDMI routing
• Added a power button

By this point, soldering experience had accumulated and the main issues were resolved.

Second Version Results

• Single-core CPU running at 60 MHz
• Linux boots successfully
• Display works, but green artifacts appear at 1280x720 resolution
• Instability when connecting USB devices
• After six months, the board degraded (memory and video problems after heating)

Third Version (Current)

The decision was made to switch to BGA components for a more powerful system.

Component Selection

• FPGA — Efinix Ti60F256 (256 BGA pins, 0.8 mm pitch)
• Memory — IM8G16D3FFBG DDR3L 1 GB (96 BGA pins)
• Board — 6-layer PCB (approximately $100)

PCB Design

DDR3 requirements:

• Data trace length matching
• Specific impedance requirements
• Numerous power decoupling capacitors (0603 minimum size)
• Separate layer allocation for address/command signals

Additional components:

• TMDS serializer (TFP410) to eliminate video artifacts
• USB current limiter
• SD card voltage switching (3.3V/1.8V for UHS-1)
• RTC with battery backup
• ESP32 module for WiFi
• Second USB-C port with USB Power Delivery support

BGA Soldering

The process of soldering the new components using a bottom heater:

• Applying solder paste through a stencil
• Melting at 221°C to form solder balls
• Soldering the bottom side with hot air at 183°C
• Hand-soldering connectors

The first attempt was successful — all contacts soldered correctly with minimal bridging.

System On Chip

The architecture includes several IP blocks arranged around the processor core and peripherals.

Third Version Results

• Processor runs at 220 MHz (compared to 60 MHz previously)
• Performance — 495 DMIPS (2.25 DMIPS/MHz) on the Dhrystone benchmark
• Display — 1920x1080 at 50 Hz without artifacts
• Memory — 1 GB DDR3, running at 333 MHz (instead of the target 400 MHz)
• Audio — acceptable quality
• SD card — SDR50 mode works stably
• WiFi — not yet tested

Future Plans

I intend to use the remaining 10K logic elements of the FPGA to create a simple GPU and attempt to run DOOM or Quake.

Key Lessons

• Proper power supply planning is critical for stability
• Test firmware should be written before designing the board
• Studying datasheets and standards (DDR3, HDMI, USB) is essential
• BGA soldering during production is simpler than repairs
• Iterative board design refinement yields the best results