Deepgrid Semi
DG32 · LOCKSTEP RISC-V MOTOR-CONTROL SILICONFIRST SILICON · SEP 2026
Discuss your application

Six functional groups.
One of them is frozen.

The DG32-LITE die is organised into six regions. Five are configurable per SKU. The safety core is closed until silicon test, because the part that catches a fault is the part that cannot be allowed to change.

Drag the die to rotate it. Scroll to walk the regions.

ILLUSTRATIVE MODEL · NOT A MASK LAYOUT · DRAG TO ROTATE

SAFE FROZEN

Safety core

Lockstep pair, fault latch, windowed watchdog

  • Lockstep CPU pair

    MAIN runs the application; CHECKER runs the same instructions two cycles later on mirrored inputs and bus responses.

  • Comparator and fault latch

    Compares every committed store; the first mismatch sets a sticky latch with its cause and drives FAULT_N.

  • Supervisor and windowed watchdog

    Two supply-good inputs with deglitching and reset sequencing, plus a watchdog that faults on a kick that is too early or too late.

WHY Motor control drives power electronics. A silent CPU fault can destroy a bridge; a second core that must agree bit-for-bit turns that fault into an observable signal.

MEMCONFIGURABLE

Memory & boot

64 KB ROM, 32 KB SRAM, QSPI flash

  • Boot ROM

    64 KB of mask-programmed logic that prints a banner, checks the flash image, copies it into SRAM and jumps to it.

  • Dual-port SRAM

    32 KB on 16 macros: one port for CPU and DMA data, the other dedicated to instruction fetch.

  • QSPI controller

    Maps external NOR flash and PSRAM into memory as a data window.

WHY A flash-less die keeps the 130 nm process simple. The cost is loading at boot instead of executing in place; embedded flash is on the roadmap.

MTRCONFIGURABLE

Motor drive

3-phase PWM, DShot × 4, timers

  • 3-phase PWM

    Centre-aligned complementary outputs on six gate pins, programmable dead-time up to 5.1 µs, shadow duty registers and an ADC sample trigger.

  • DShot × 4

    Encodes digital ESC frames for four channels onto the PWM pads.

  • Timers × 2

    Two 32-bit timers with compare interrupts.

WHY A firmware brake is only as fast as the loop. A bridge fault needs the safe state now, so the brake path goes straight to the gates.

SNSCONFIGURABLE

Sensing & math

Encoder + Hall, SAR ADC, CORDIC

  • Encoder and Hall decode

    Quadrature A/B/Z with index and three Hall inputs, with edge timestamps.

  • SAR ADC

    8-bit differential converter at ~200 kSa/s, fired by the PWM.

  • CORDIC

    sin/cos, atan2 and magnitude in a fixed 53–58 cycles per operation.

WHY An external SPI ADC read costs ~1.9 µs of a ~5 µs loop, the largest single cost. On-die it becomes a start pulse and a poll.

COMCONFIGURABLE

Connectivity

UART × 2, SPI, I²C, QSPI, GPIO

  • UART × 2

    UART0 carries the boot banner and console; UART1 is free for telemetry.

  • SPI master

    Up to 25 MHz, all four modes, 0.64 µs per 16-bit word.

  • I²C master

    7-bit addressing at 100 or 400 kHz with clock stretching and true open-drain pads.

  • GPIO

    General-purpose pins with synchronised inputs and separate set and clear registers.

WHY The set a motor drive needs, done in silicon where timing matters. CAN-FD is a roadmap item.

SYSCONFIGURABLE

Bus, system & test

Two-master bus, interrupts, DMA, JTAG

  • On-chip bus and error slave

    A lightweight AXI-style bus: two masters (CPU first, then DMA), one transaction at a time, every peripheral a 32-bit register slave.

  • Interrupt controller

    16 sources plus a software interrupt, delivered identically to both cores.

  • DMA

    A second bus master that moves memory blocks without CPU copy loops.

  • System control

    Part ID, pin multiplexing and eight software clock gates for idle peripherals.

  • Test access

    A JTAG port reaching 13 internal scan chains for production test.

WHY There is no debugger halt on this die. A hung bus would be a brick, so every bad access has to be visible and survivable.

SAFE FROZEN

Five of the six are yours.
This one is not.

Motor control drives power electronics. A silent CPU fault can destroy a bridge; a second core that must agree bit-for-bit turns that fault into an observable signal.

  • Lockstep CPU pairMAIN runs the application; CHECKER runs the same instructions two cycles later on mirrored inputs and bus responses.
  • Comparator and fault latchCompares every committed store; the first mismatch sets a sticky latch with its cause and drives FAULT_N.
  • Supervisor and windowed watchdogTwo supply-good inputs with deglitching and reset sequencing, plus a watchdog that faults on a kick that is too early or too late.

Pre-silicon. The die image is an illustrative model, not a mask layout, and the figures above are design values verified in simulation and static timing.