Track · RTL → GDSII · Interactive

From Verilog to a real chip

On an FPGA, your design is loaded into a chip that already exists. On an ASIC, your design becomes the chip. The same Verilog goes through a different flow and comes out as GDSII: the layer-by-layer drawing a foundry turns into silicon. Since 2020, every step can be done with free tools and an open process kit.

VIDEO LESSONS · COMING SOON This page is interactive now; voiced lessons for this track are in planning.

FPGA flow vs. ASIC flow

RTL → FPGA (Tiers 0–5)RTL → GDSII (this track)
Synthesis maps logic toLUT4s, ALU carry cells, flip-flops, block RAMStandard cells from a PDK library (NAND2, XOR2, DFF, buffers…)
Placement puts cellsOn fixed sites in the FPGA's tile gridOn rows you define in a floorplan you choose
Clock networkBuilt into the FPGAYou build it: clock tree synthesis
RoutingProgrammable switches between fixed wiresReal metal wires drawn on 5–6 metal layers
OutputA bitstream (.fs), loaded in secondsA GDSII file, manufactured over months
A mistake costsA rebuildA respin: a new tape-out
Open toolsYosys, nextpnr, Apicula, openFPGALoaderYosys, OpenROAD, Magic, KLayout, Netgen, OpenSTA (packaged as OpenLane)

Walk the flow

A small block of about 150 standard cells, drawn for illustration, goes through each step of an OpenLane-style flow on an open PDK (SKY130-like). Step through it; at the end, switch layers on and off.

BENCH · RTL TO GDSII

Reads → writes
In the flow

  
Watch for

What a standard cell is

An FPGA's basic block is a LUT: a tiny memory. An ASIC's is a standard cell: a hand-drawn layout of a few transistors that does one job, such as a NAND2, an XOR2 or a flip-flop, in several drive strengths.

Every cell in a library has the same height, so cells can sit side by side in rows with their power rails lined up. In SKY130's high-density library (sky130_fd_sc_hd) that height is 2.72 µm.

The library comes with timing models (.lib), abstract shapes for placement (.lef) and full layouts (.gds). Synthesis only needs the first; the rest of the flow uses all three.

Video lessons in this track

  1. G.1Standard cells and the PDK. From LUTs to cells: what SKY130 gives you, and synthesising NandTier's ALU to it.COMING SOON
  2. G.2Floorplan and placement. Die size, power rings, rows, and where each cell lands.COMING SOON
  3. G.3Clock tree and routing. A balanced clock, then real metal on five layers.COMING SOON
  4. G.4Sign-off and tape-out. DRC, LVS, timing, GDSII, and a Tiny Tapeout tile.COMING SOON

Try it for real

START HERE

Tiny Tapeout

A shared shuttle: many small designs on one chip, each in a tile of roughly 1,000 standard cells. Its GitHub template runs the whole flow in GitHub Actions and shows your layout in the browser, with no install. tinytapeout.com

FROM NANDTIER

What fits in a tile

NandTier's UART, a PWM, or the ALU fit in a tile. The whole CPU would need several. Start small: the same Verilog you simulated in Tier 1 goes in unchanged.

VERIFY FIRST

Testing matters more

A bug on an FPGA costs a rebuild; on silicon, a respin. Simulation, cocotb and formal proofs from the toolbox lessons are how chip teams avoid that. Tiny Tapeout projects ship with a cocotb test.

A minimal OpenLane 2 configuration

{
  "DESIGN_NAME": "nt_uart",
  "VERILOG_FILES": "dir::src/*.v",
  "CLOCK_PORT": "clk",
  "CLOCK_PERIOD": 40.04
}
// run it:  openlane config.json
// 40.04 ns = NandTier's 24.975 MHz system clock