72 TOPS on One Board: The Cockpit Domain Controller PCB Challenge

A car that sells for around 100,000 yuan now ships with a computing module rated at 72 TOPS. That figure, once the preserve of dedicated autonomous driving hardware, is increasingly delivered by a single board that consolidates cockpit infotainment and driving assistance into one domain controller. The architectural intent is straightforward: fewer boxes, fewer wiring harnesses, more shared computation.

The manufacturing consequence is not straightforward at all. Consolidating functions means consolidating heat, current, and signal complexity onto one substrate, in a product that must survive fifteen years of thermal cycling, vibration, and humidity while meeting automotive quality expectations. Understanding what that does to the board explains why the automotive PCB conversation is shifting from piece count to capability per panel.

From Many ECUs to a Few Domain Boards

Conventional vehicle electronics distributed function across many electronic control units, each with its own housing, connector, and relatively modest circuit board. Cabin-driving integration collapses several of those functions into one controller with a shared system on chip, or a small cluster of high-performance devices.

Two effects follow. The first is a change in PCB value distribution. The total number of boards in a vehicle may not fall dramatically, but the value and difficulty migrate from many simple boards to fewer, far more demanding ones. The second is that reliability expectations do not consolidate along with the architecture. A domain controller that fails takes down several vehicle functions at once, whereas a distributed architecture degraded more gracefully. That asymmetry is why integrated controllers face stricter validation even when their individual components are well understood.

For the PCB, integration means high-speed interfaces between the processor and memory, display links to multiple screens, camera and sensor inputs, power delivery for tens of watts of dynamic load, and an RF front end for connectivity, all sharing one stackup, one return path system, and one thermal solution.

Why Cabin-Driving Fusion Changes the Board

Pin density and escape routing. High-performance automotive processors and their companion memory use fine-pitch ball grid array packages with high pin counts. Escaping those pins from a limited area requires narrow lines and microvia structures, which is why automotive designs increasingly resemble advanced consumer electronics in their inner layers while retaining automotive materials and qualification.

High-speed channels. Display serial links, camera interfaces, memory buses, and Ethernet all impose controlled impedance requirements and loss budgets. With several interfaces running simultaneously, crosstalk and return path management become the dominant layout constraints rather than a secondary concern. This is a design conversation that benefits from early engagement with fabrication capability, because line width, dielectric thickness, and stackup symmetry determine what the layout can achieve. Teams that treat PCB design and layout limits as fixed inputs tend to discover them late.

Mixed signal domains. The same board carries RF for connectivity, high-speed digital for computation, and switching power conversion that produces broadband noise. Physical separation, ground partitioning, and careful placement of switching nodes determine whether the radio receives clean spectrum and whether the memory bus meets its eye mask.

Layer count. Signal layer demand, reference planes, and power distribution together push these boards into the ten to twenty layer range, with some designs going higher. Each additional pair raises cost and increases the difficulty of holding registration and impedance across the panel.

Thermal and Power: The Hardest Constraint

A 72 TOPS module in an enclosed automotive housing, operating in a cabin that can reach 85 degrees Celsius in the sun, has very little thermal headroom. Two situations dominate the thermal design, and only one of them is steady state.

The steady-state case is the easier one. Continuous computing load produces a fairly predictable heat flux that can be managed with copper area, thermal vias, and conduction to a housing heatsink. The difficult case is transient: a burst of camera and fusion processing, or a software update, can drive power far above the average for a short interval. Junction temperature responds to the peak, not the average, so the board has to conduct heat away fast enough to keep the die inside its limit during the burst rather than only in equilibrium.

This is where laminate choice and copper architecture matter together. Thermal vias under the processor need adequate diameter and barrel plating to conduct effectively, and they must be arranged so that the underlying copper spreader distributes heat without creating a hot spot. Thick copper helps where current is high, but it also changes lamination behaviour and etched feature tolerance, so it is a manufacturing decision rather than a simple upgrade.

Power delivery adds its own constraint. Automotive processors demand large transient current with tight voltage tolerance, which means a dense network of decoupling capacitors close to the load, a low-impedance power plane, and enough copper to carry the current without significant drop. Every capacitor needs a low-inductance path to the planes, and every plane needs to be partitioned so that switching noise does not couple into sensitive analogue or RF sections.

Materials and Layer Count for Automotive HDI

Automotive electronics have traditionally favoured FR-4 with high glass transition temperature and proven thermal cycling behaviour, because qualification history is worth something when a product must survive fifteen years in the field. That foundation remains valid, but integration pushes it in two directions.

Where signal speed is high enough, low-loss laminates earn their place on the high-speed channels, even if the rest of the stackup stays conventional. Selective use of low-loss material in a hybrid stackup is a legitimate engineering approach: it addresses the loss budget where it matters without paying for advanced dielectric on layers that carry only power and low-speed signals. Getting a hybrid stackup right requires the fabricator to control multiple materials in one lamination cycle, which is a capability question rather than a specification question, and one worth confirming in a PCB capability review before the design is frozen.

HDI structures become necessary rather than optional. Microvias, buried vias, and via-in-pad allow the escape density that fine-pitch packages demand, and they shorten the return paths that high-speed channels require. On the manufacturing side, higher layer counts increase the number of lamination cycles, each of which is an opportunity for registration drift or resin starvation. A twenty-layer automotive HDI board may pass through four or more pressing operations, and the cumulative effect on layer-to-layer alignment determines whether the finished board meets its impedance targets.

Automotive qualification adds a further layer. Component and process requirements derived from automotive standards govern the temperature range, and the expectation is typically minus 40 to plus 125 degrees Celsius at the board level, with vibration and thermal shock added for modules mounted away from the cabin. Board materials must remain dimensionally stable through those excursions, which is why the mechanical properties of the laminate, not just its electrical ones, drive selection.

Reproducibility Is the Real Automotive Requirement

An automotive programme does not buy one board. It buys several hundred thousand boards a year, produced over a decade, from a process that must not drift. The engineering achievement is not producing a working prototype; it is producing the same board ten thousand times.

That reframes what a manufacturer needs to demonstrate. Process windows must be wide enough to absorb normal material variation. Soldering and assembly must be qualified for the automotive temperature range, with inspection and electrical test providing objective evidence rather than a visual sign-off, which is why a defined PCBA test strategy is part of the design rather than an afterthought. Traceability must connect a finished controller to the laminate lot and process parameters that produced it, so that an anomaly in the field can be bounded quickly. These are process disciplines, and they belong in a quality management system that predates the first automotive order rather than being assembled in response to one.

Volume capability matters for the same reason. A supplier that can build ten boards well but cannot scale to hundreds of thousands without changing its process has not solved the customer’s problem. High-volume PCB assembly capability, combined with controlled PCB fabrication, is what allows a cockpit domain controller programme to move from sample to series production without requalification.

The broader trend is clear. Cabin-driving integration does not necessarily change how many boards a vehicle contains. It changes how much computation, communication, and power each critical board carries, and whether that complexity can be reproduced reliably at automotive volumes. When that architecture reaches mainstream vehicle platforms, the variable worth watching will not be board count per car, but how much function each core board integrates and whether the manufacturing system behind it can hold that complexity stable.

Frequently Asked Questions

Why does cabin-driving fusion increase PCB difficulty? It consolidates multiple functions onto one substrate, so high-speed digital, RF, power conversion, and thermal management constraints must all be satisfied simultaneously in a single stackup.

Do automotive domain controllers need HDI? Most current designs do, because fine-pitch processor packages and multiple high-speed interfaces need microvia escape routing and short return paths. The layer count typically lands in the ten to twenty range.

Is low-loss laminate necessary for automotive boards? Only where the loss budget requires it. Hybrid stackups that use low-loss material on high-speed layers and conventional high-Tg material elsewhere are a common and cost-effective compromise.

What limits thermal performance more, steady state or transients? Transients. Junction temperature responds to peak power, so the board must conduct heat away quickly during short bursts rather than only reaching equilibrium under average load.

What should an automotive customer verify in a PCB partner? Process capability and maturity, automotive-grade materials with stable supply, test and inspection strategy, lot-level traceability, and demonstrated volume production rather than prototype skill alone.