Get the newsletter
Nvidia · Automotive · Filed May 13, 2025 · USPTO published Nov 19, 2026 · This week's release

Nvidia Patents a Car Radar Chip That Processes Data Before Storing It

A car with five radar sensors can flood its central computer with around 16 gigabits of raw data every second. Nvidia's patent filing suggests skipping the storage step entirely and doing the first round of math as the data comes in.

A car equipped with radar sensors at each corner connected to a central onboard computer that handles incoming sensor signals. Drawing from patent filing US 2026/0367220 A1.
A car equipped with radar sensors at each corner connected to a central onboard computer that handles incoming sensor signals.All 11 drawings ↓
Publication number
US 2026/0367220 A1(US20260367220A1)
Applicant
NVIDIA Corporation
Filed · Published
May 13, 2025 · Nov 19, 2026
Inventors
Romain Ygnace, Thomas Joseph Tomazin
Status when we published
Waiting for an examiner
Document
20 claims
US classification
342/70
Examiner
PERVIN, NUZHAT (Art Unit 3648)

What Nvidia's radar-first receiver actually does

Every time a car with driver-assist features checks the road around it, its radar sensors fire off bursts of signals and send back the echoes. That is a lot of raw data. One sensor with four receiving channels can produce about 3.2 gigabits every second, and a car with five of them pushes roughly 16 gigabits toward its main computer.

In the usual setup, the main computer first dumps all of that raw data into memory, then pulls it back out to do the math, then writes the results to memory again. That is a lot of back-and-forth, and it clogs the memory and burns power.

Nvidia's filing describes a different order of operations. A small processing block sits right at the door of the main computer's radar receiver. It does the first pass of math on each burst of echoes before anything is saved, so only the finished, much more useful results ever reach memory.

From the filing · CLAIM 1
… radar processing circuitry to process the radar data into a first range fast Fourier transform (FFT); and one or more direct memory access (DMA) engines to write the first range FFT into the memory.

Translation: Specialized hardware converts the radar signals into a math format before saving them.

How the chip turns radar echoes into distances

The patent covers a radar receiver inside a car's central compute cluster (the main computer that runs driver-assist software). Radar sensors around the vehicle send raw samples to this receiver over a connection such as Ethernet or a camera-style data link.

Instead of writing that raw data to memory, the receiver holds it briefly in a small intermediate buffer (a temporary holding area, described as anywhere from about 4 KB to 8 MB). Once the buffer has every sample from one radar burst (called a chirp), built-in radar logic runs a range FFT (a Fourier transform, a math step that turns raw echoes into how far away each object is). A DMA engine (a dedicated data mover that copies information into memory without bothering the main processor) then writes only that result to memory.

Later, once a full radar cycle is done (the description gives examples of 256, 512 or 1,024 chirps), the receiver pulls the range results back out and runs a second transform, a Doppler FFT, which reveals how fast objects are moving. That is written back to memory for the driver-assist software to use.

  • Raw data is held only briefly, then discarded.
  • Several sensors can share one buffer, sorted by the labels in the data stream.
  • The receiver can be a single chip, a custom chip, or a reprogrammable one.

Why this matters for crowded car computers

For you as a driver, the benefit would be indirect: less data clogging the car's computer can mean faster reaction times and lower power draw, which matter more as cars add more and sharper radar sensors. The filing says the main computer, not each sensor, is where cooling and space are easier to manage, so moving work there is attractive.

For Nvidia, this fits its role as a supplier of the computers inside driver-assist systems. A chip block that makes radar cheaper to process in a central computer is a selling point for automakers consolidating their electronics. The filing also lists aircraft, trains, ships and factory machinery as other places the design could work.

That makes this Nvidia's 115th filing we've tracked since May in our self-driving sensing race watch, following its work on blocked sensor detection and blink-based drowsy driver cameras.

Source: US 2026/0367220 A1 at the USPTO (PDF). Full patent text and figures are in the official publication.

Get our take in your Top Stories

Liked this breakdown? Add Patentlyze as a preferred source on Google, and our plain-English take shows up more often in your Top Stories the next time Nvidia patent news breaks.

There are more where this came from

We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.

The drawings · 11 sheets

11 drawing sheets from US 2026/0367220 A1 · click any drawing to enlarge

How the receiver chip analyzes radar signals before saving them, drawing sheet 1 from US 2026/0367220 A1
Sheet 1 How the receiver chip analyzes radar signals before saving them
Sensors around a car feed into a central computer, drawing sheet 2 from US 2026/0367220 A1
Sheet 2 Sensors around a car feed into a central computer
Staggering radar sampling and calculations to keep data moving, drawing sheet 3 from US 2026/0367220 A1
Sheet 3 Staggering radar sampling and calculations to keep data moving
Turning raw radar echoes into distance readings step by step, drawing sheet 4 from US 2026/0367220 A1
Sheet 4 Turning raw radar echoes into distance readings step by step
Calculating vehicle distance without storing raw radar pulses, drawing sheet 5 from US 2026/0367220 A1
Sheet 5 Calculating vehicle distance without storing raw radar pulses
Parts connecting radar sensors to the car chip, drawing sheet 6 from US 2026/0367220 A1
Sheet 6 Parts connecting radar sensors to the car chip
Sending multi-level digital signals across a physical link, drawing sheet 7 from US 2026/0367220 A1
Sheet 7 Sending multi-level digital signals across a physical link
Computer parts that can run the automated driving software, drawing sheet 8 from US 2026/0367220 A1
Sheet 8 Computer parts that can run the automated driving software
Show the other 3 sheets
Connecting multiple superchips to share car sensor workloads, drawing sheet 9 from US 2026/0367220 A1
Sheet 9 Connecting multiple superchips to share car sensor workloads
Combining the main processor and graphics on one chip, drawing sheet 10 from US 2026/0367220 A1
Sheet 10 Combining the main processor and graphics on one chip
Linking dozens of specialized AI chips for heavy data processing, drawing sheet 11 from US 2026/0367220 A1
Sheet 11 Linking dozens of specialized AI chips for heavy data processing
Patent filing page

Watchlists this filing belongs to

All watchlists →
Reader comments

No comments yet

Start the discussion

Real name or a handle, either is fine. Comments are read by a person before they appear, so allow a little time. Keep it about the filing.