What Would Support for Trustworthy AI Inference Be Worth on Your Microcontrollers?
Your customers want a confidence number with every inference. Today they get one by running the same model several times over and reading the disagreement between the runs. That multiplies the work by the number of runs, on parts with neither the headroom nor the power budget for it. UxHw® computes with distributions, so one pass returns the spread. It reaches your customers through your compiler toolchain, developer tools, development environments, and AI tools. Pick your company below and set the shipment uplift you think that earns.
Your CompanyThe segment each company reports that contains its microcontrollers, quoted as the company writes it. Only two of the nine print a microcontroller figure, so most of these lines also carry other products and the base is generous. Sources and what sits inside each line are in market/model-mcuVendors.md.
Shipment Uplift, %How much more of your microcontroller line you would ship with UxHw in the cores, supported across your compiler toolchain, your developer tools, your development environments, and your AI tools. Nobody has measured that, so it is your figure and not a Signaloid forecast. Everything in the money view scales with it.
10
Your Own Segment RevenueOverrides the list. Use it if your microcontroller revenue differs from the segment your company reports, which it will, or if your company is not on the list.
Models in the EnsembleHow many members you run to get a calibrated answer, or how many forward passes Monte Carlo (MC) dropout takes. The published benchmark is five branches, so five is the like-for-like setting.
5
Inferences Per Device Per DayOne every ten seconds is 8,640 a day. This drives energy rather than cost, because the device is already bought.
8,640
Fleet SizeDevices in the field. On a new design, this is the annual build.20,000
Signaloid Moduleη is the fidelity setting, the detail the processor keeps in each distribution. Higher is more faithful and slower. Every module has a ceiling. The ceiling matters more than price here. The published figure is measured at η = 32, which C0-microSD cannot reach, so the $49 part is the wrong part for this workflow.
Avoided RespinBoard redesign you avoid by not needing new silicon on the printed circuit board. Typically $50k to $250k for a moderately complex embedded board.150,000
Avoided RecertificationEmissions and safety testing a redesigned board has to repeat. A module in an existing slot changes no enclosure and no radio, so it triggers none of it.
60,000
Months of Delay AvoidedHow long a redesign would hold the product up, from layout to qualified parts in hand.
12
Cost of a Month of DelayWhat a month of not shipping costs you. Set it to zero to price the redesign on engineering alone.
80,000
Cost of Fitting One ModuleFitting is not free even when the module is. At $99 a card a technician visit can cost more than twice the hardware. Leave it at the first option only if the work truly rides along with something you already do.
Host Processor CostWhat the processor running the ensemble costs you per unit. The default follows the scenario you picked, a Raspberry Pi 5 [1, 2] or Compute Module 5 [3] at an Approved Reseller list price, read August 2nd 2026. Your own supplier agreement will differ.
Host Processor PowerPower drawn while running all members. The default follows the scenario you picked, a floor taken from the Raspberry Pi Compute Module 5 datasheet operating current, read August 2nd 2026. Raspberry Pi publishes no under-load figure for the Pi 5 itself, so this understates it.
Battery CapacityUsed only to express the power difference as runtime, the unit you budget in.
Override the speedup
This calculator's figures rest on one measured multiple, cited
with its baseline under the result. If you have measured your own, on your own workload, enter it
here and the arithmetic follows it. Nothing leaves the page. The citation under the figures still
reports what Signaloid measured, a banner above the result says the numbers are yours, and the
exported card and the quote mail both record the substitution. Clear the field to go back.
There is no single multiple to override here. This
calculator's figures rest on several separately measured quantities rather than on one speedup.
Each is cited with its own baseline under the result. Overriding one of them alone would give
a figure that follows from no measurement, so the field the other calculators carry is absent.
Speedup, ×
C0-microSD
$49, shipping today from Mouser. 12 MHz, 128 KiB, 17.1 mW average, up to η = 16.
$49, shipping today from Mouser. 12 MHz, 128 KiB, 17.1 mW average, up to η = 16.
C0-microSD+
$99, engineering samples available today, contact sales@signaloid.com. At least 60 MHz, 320 KiB, up to η = 64.
$99, engineering samples available today, contact sales@signaloid.com. At least 60 MHz, 320 KiB, up to η = 64.
C0-SD
$199, engineering samples available today, contact sales@signaloid.com. 90 MHz, 448 KiB plus 128 MiB PSRAM, up to η = 128.
$199, engineering samples available today, contact sales@signaloid.com. 90 MHz, 448 KiB plus 128 MiB PSRAM, up to η = 128.
Benefits