Skip to content

Host Controller Selection

The main program of this project requires a pretty powerful processor. With an ideal image refresh rate between 60-120Hz, here's the calculations for maximum processor throughput and transfer speeds:

Per-Frame Calculations - 6 panels - 64 x 64 LEDs per panel - 8 bits color per LED

    6 * 64 * 64 = 24,576 LEDs
    8 * 3 colors = 24 bits = 3 bytes
    24576 * 3B = 73728 B = 72 KB per full frame

    73 KB * 120 FPS = 8,847,360 B / s = 8.8MB /s

This transfer speed is able to happen with a fast parallel bus and DMA or a memory connection.

The processor needs to be able to calculate frames at or above 120Hz, which means calculating 24,576 pixel values 120 times/s. This leads to 2,949,120 operations / sec. Simple operations such as just pixel displaying is easy, negligible performance use. However simulation and calculation, such as a 6-sided digital sand simulation, would require ~3MFLOPS (Mega Floating point Operations Per Second).

STM32 Processor Comparison for LED Cube Project

Compared Families

  • STM32N6x7
  • STM32N6x5
  • STM32H7S3
  • STM32H7S7
  • STM32H745
  • STM32H747
  • STM32H562
  • STM32H563

High-Level Architectural Comparison

Feature N6x7 N6x5 H7S3 H7S7 H745 H747 H562 H563
Core Cortex-M55 Cortex-M55 Cortex-M7 Cortex-M7 M7 + M4 M7 + M4 Cortex-M33 Cortex-M33
Max Clock ~800 MHz ~800 MHz ~600 MHz ~600 MHz 480 / 240 MHz 480 / 240 MHz ~250 MHz ~250 MHz
SIMD / Vector Helium (MVE) Helium (MVE) DSP only DSP only DSP only DSP only Basic DSP Basic DSP
FPU Yes Yes Yes Yes Yes Yes Yes Yes
Neural Accelerator Yes (Neural-ART NPU) Yes (Neural-ART NPU) No No No No No No
GPU NeoChrom NeoChrom Chrom-ART (DMA2D) NeoChrom Chrom-ART Chrom-ART None None
ISP / Camera Yes Yes Limited Limited Limited Limited No No
Internal SRAM ~4.2 MB ~4.2 MB ~620 KB ~620 KB ~1 MB ~1 MB ~640 KB ~640 KB
Internal Flash External required External required Boot flash only Boot flash only Up to 2 MB Up to 2 MB 2 MB 2 MB
Cache I + D cache I + D cache L1 cache L1 cache L1 cache L1 cache Cache Cache
External Memory FMC, XSPI, HexaSPI FMC, XSPI, HexaSPI FMC, xSPI FMC, xSPI FMC, QSPI FMC, QSPI OctoSPI OctoSPI
Security Advanced + TrustZone Advanced + TrustZone Advanced Advanced Standard Standard TrustZone TrustZone

Performance & Capability Analysis (Project-Relevant)

STM32N6x7 / N6x5

Strengths: - Fastest CPU (800 MHz) - Helium vector extensions (major boost for math-heavy simulation) - 4.2 MB contiguous SRAM - Dedicated Neural-ART NPU (~600 GOPS class) - NeoChrom GPU - Modern memory subsystem

Impact on Cube: - Excellent for 3D particle simulation - Large simulation state objects - Multi-buffer rendering - Future ML integration - UI acceleration

Weakness: - Requires external flash - Higher system complexity


STM32H7S7 / H7S3

Strengths: - Very strong Cortex-M7 core - NeoChrom GPU (H7S7) - Mature ecosystem - Good real-time determinism

Limitations: - Much less SRAM (~620 KB) - No AI accelerator - No Helium vector math

Impact: - Capable of 2D/3D simulation - Requires tighter memory management


STM32H745 / H747

Strengths: - Dual-core (M7 + M4) - Good for task partitioning - Up to 2 MB flash

Limitations: - Lower max clock - No AI accelerator - Much smaller RAM than N6

Impact: - Structured multi-core systems - Less ideal for large simulation buffers


STM32H562 / H563

Strengths: - Secure - Efficient - Good general MCU

Limitations: - 250 MHz max - Limited RAM - No graphics accelerator - No ML hardware - No vector extensions

Impact: - Not recommended for high-performance simulation workloads


Vector Math Importance

Cortex-M55 (Helium): - SIMD acceleration - Vectorized float math - Faster matrix transforms - Faster particle integration loops

Cortex-M7: - Scalar floating-point - Strong but not vectorized

For heavy particle systems: M55 + Helium is a major architectural advantage.


Practical Recommendation

MCU Line Verdict
STM32N6x7 / N6x5 Best overall choice. Most headroom and future expansion capability.
STM32H7S7 Strong alternative if avoiding N6 complexity.
STM32H7S3 Good but less graphics acceleration.
STM32H745 / H747 Good dual-core option but less memory headroom.
STM32H562 / H563 Not recommended for this application.

Some of this was formatted or sourced using ChatGPT

N647B0 has been chosen based on given factors

The high frequency and high RAM capacity, coupled with specialized hardware accelerators, it adds a lot of flexibility and expansion for the future of th eproject

However, a major issue arises - lack of onboard flash.