← TinyWhisper documentationGenerated 2026-11-15 from commit 10ef5004 by ihp130/scripts/gen_dashboard.py

TinyWhisper chip dashboard

IHP SG13G2 130 nm · 2 × 2 mm WSPR short-wave transmitter · taped out March 2026 (SoC4219) · data from the committed repo outputs

— of — sign-off checks pass outright across chip, RISC-V and IQ modulator. Timing closes at every corner with — ns worst setup slack on the core. The analog risk is filter corner spread: — % of 200 Monte Carlo samples meet fc ≥ 350 kHz. The analog block draws — from the 1.5 V supply in the post-layout transient. Silicon is back but not measured yet.

RV32E core→30-bit NCO + CORDIC (I/Q)→ΔΣ modulator, order 1/2, OSR 32–256→SPDT select→3rd-order MFB LPF, 400 kHz→Passive TG mixer, 25 % LO→Off-chip BPF + PA
Core setup slack, worst corner—ns17 ns clock (56 MHz), 0 setup or hold violations
Core power, typ 1.5 V 25 °C—mWPost-PnR estimate, riscv_top only
Core std cells—Yosys stat after synthesis
LPF corner, nominal—tt, 27 °C, 1.5 V ngspice
LPF fc ≥ 350 kHz yield—%CACE, 200 samples
RF output, 10 m band—dBm— mVp (— mVrms), — µW into 1 kΩ after the 28 MHz BPF, PEX tran

Analog: IQ modulator (iqmod_top, 337 × 142 µm)

Per channel: SPDT input select, fully differential 3rd-order Butterworth MFB low-pass with Barthelemy/Manfredini inverter OTA, passive transmission-gate mixer.

MFB filter and OTA magnitude response

The filter corner sits at 400 kHz and the 56 MHz clock image is pushed far down. Dashed lines mark the 400 kHz target and the 56 MHz clock.

MFB LPF closed loopOTA open loop

Monte Carlo spread of the filter corner

200 CACE samples. Bars below the 350 kHz limit are red.

CACE specs over 200 samples

BlockSpecLimitMinMeanMaxPass %Status

Nominal simulation results

BlockMetricValueData file

Digital: RISC-V SoC (riscv_top, 700 × 1130 µm)

Multi-cycle RV32E + Zicsr core, 16 × 32-bit registers, code in external SPI SRAM. UART 115200, I²C 100 kHz, 4 + 4 GPIO, timer, external and UART interrupts. DSP chain: 30-bit NCO, 30-iteration CORDIC, ΔΣ modulator, 4-phase LO divider.

Post-PnR slack per corner (ns)

Setup WNSHold WNS

Power per corner, by group (mW)

SequentialCombinationalClock

Standard-cell area by function (µm²)

Flip-flops take the most area: the CORDIC, NCO, ΔΣ and peripheral registers are 30 bits wide.

RTL size by file

— lines in 20 files. memory.sv holds the MMIO map at 0x800000.

CPU / SoCDSPPeripheralSim only

System and sign-off

WSPR band plan, chip-level facts, and the state of every verification step.

WSPR band plan

RF = LO + IF. The NCO generates the IF tone. The LO is the 56 MHz clock divided down.

BandRF (MHz)LO (MHz)IF (kHz)LO div

Chip facts

ItemValue

Sign-off status

BlockCheckToolResultStatusReport

Known errata (ihp130/doc/errata.md)

Example: TW-E01, sram_ce is active high but the APS6404L SRAM on the PCB needs active low.

Definitions and caveats

WNS is the worst negative slack. A positive value means timing is met. Yield is the share of CACE samples whose -3 dB corner is at or above 350 kHz. CACE ran at tt and 27 °C with a 1.2 to 1.8 V supply sweep. Process corners and temperature are commented out in the yaml, so this is not a full PVT sweep. Core power excludes the pads and the analog macro. RF output power is the average of v(vout_RF_filt)² / 1 kΩ over 8 µs of the committed iqmod_top_tb_tran waveform (PEX netlist, tt, 1.5 V, 28 MHz LO), behind the 1 kΩ / 1 kΩ coupled-resonator band-pass. That waveform is from the earlier Magic PEX run. dBm is 10·log10(P / 1 mW) at 1 kΩ. Vp is the peak of a single tone with the same power, √2 × Vrms. The largest sample in the window is — mV, a little higher because the filtered output is not a pure tone. The post-layout supply power and RF rows come from the KPEX 0.9.0 netlist with its parasitic resistors shorted (PR 24), because the full-RC transient over 16 µs would take days. No lab measurement exists yet.