Opening: Tell Us About Your Program

• What is the program name, the OEM (if applicable), and your target Start‑of‑Production (SoP) year?
• Which organization best describes your role on this program? Options: OEM - Autonomous/ADAS division, Tier‑1 ADAS supplier, Tier‑1 system integrator, Automotive software house, Other
• Which sensors are in scope for the baseline architecture? Options: Cameras, Radar, Lidar, Ultrasonic, IMU/GPS, Other
• Who is the primary technical owner(s) we should engage with (titles/teams)?
• What is the program’s high‑level performance priority among these options? Options: Max inference throughput, Lowest latency/determinism, Lowest power for thermal budget, Software flexibility and lifecycle, Cost and integration simplicity

Are You Building for Tomorrow or Yesterday?

• If your compute platform were frozen today, which planned AV/ADAS features would you have to drop in three years?
• Which compute bottleneck hurts you most right now? Options: NN inference throughput, Real‑time CPU scheduling, Memory bandwidth, PCIe/ethernet I/O, Thermal throttling, Other
• Give a concrete recent example where compute limits forced a design trade — what did you change and why?
• How long has that bottleneck been constraining your roadmap? Options: Less than 6 months, 6–12 months, 1–2 years, More than 2 years
• Which network families matter most for your evaluation (pick all that apply)? Options: ResNet/Backbones (classification/feature), SSD/YOLO/Detection, Transformer-based perception, Segmentation (DeepLab, UNet), Point‑cloud networks (PV‑RCNN, PointPillars), Custom proprietary nets
• What target inference throughput or TOPS do you expect for your critical networks (sustained, per SoC)? Options: <10 TOPS, 10–50 TOPS, 50–200 TOPS, >200 TOPS, Unsure / need guidance

Where the Heat and Power Bite

• If thermal or power caps forced you to reduce model fidelity, how would that impact safety margins or user features?
• What is the vehicle or module sustained thermal budget for compute (watts)? Options: <10 W, 10–25 W, 25–75 W, 75–150 W, >150 W, TBD/Varies by variant
• Do you need strict sustained power, or are short high‑power peaks acceptable with duty cycling? Options: Sustained only, Peaks OK with cooling strategy, Peaks tolerated but limited, Unsure — want recommendations
• Describe any mechanical or cooling constraints we should know about (e.g., sealed enclosures, limited airflow, chassis locations).
• How sensitive is your OEM to package thermal limits versus overall vehicle thermal budget? Options: Very sensitive to package, More sensitive to vehicle-level budget, Both equally, Unsure

When Real‑Time Fails You

• Tell us about a time latency or unpredictable jitter caused a missed detection or degraded control — what was the consequence?
• What are your hard real‑time latency targets (end‑to‑end) for safety‑critical paths? Options: <5 ms, 5–20 ms, 20–50 ms, 50–100 ms, >100 ms, Varies by function
• What worst‑case latency or jitter do you need to prove in verification to meet acceptance? Options: Deterministic bounded microseconds, Bounded milliseconds, Statistical worst‑case with percentiles, Unsure—need guidance
• Who currently owns deterministic scheduling and real‑time validation on your team? Options: Functional safety engineers, Real‑time software team, SoC/software architecture, Third‑party integrator, We don't have a clear owner
• What tooling or data do you have now for latency and jitter analysis (e.g., trace captures, CI benchmarks)? Options: Detailed traces and CI, Basic profiling reports, Ad‑hoc measurements only, None yet

The Safety Conversation No One Wants to Drop

• If an OEM demanded ASIL‑D evidence tomorrow, where would your current safety case be weakest?
• Which ASIL target(s) are you designing toward across functions? Options: ASIL‑A, ASIL‑B, ASIL‑C, ASIL‑D, Mixed/varies by function, Unsure
• Which artifacts do you already have that support functional safety evidence? Options: FMEA/FMECA, HARA, Safety requirements, ASIL decomposition, Verification reports, None/starting from scratch
• What verification data can you share under NDA to accelerate safety mapping (e.g., model test sets, scenario logs, CI traces)? Options: Per‑network test datasets, Closed‑loop scenario logs, Hardware‑in‑the‑loop traces, Safety test reports, None yet
• How would you describe the emotional impact of safety uncertainty on the program timeline and team morale?

Who Holds the Keys?

• If one person could veto the silicon, who would they be and what do they care about most?
• Which stakeholders will be required to sign approval gates? Options: OEM program management, OEM safety/validation, Tier‑1 architecture, Perception leads, Procurement, Other
• What weight do each of these criteria carry in the decision (pick up to three most important)? Options: Performance (throughput), Deterministic latency, Power/thermal, Safety evidence, Supply reliability, Cost
• What is your procurement timeline and key approval milestone dates we should coordinate to? Options: RFP in next 3 months, Evaluation in 3–6 months, Design freeze 6–12 months, SoP planning 1–2 years, Other
• Have you experienced supplier supply‑chain concerns in past programs, and if so, how did they affect choices?

If This Went Perfectly

• Describe the one demo or data point that would make the OEM immediately comfortable — what does it prove and to whom?
• Which acceptance criteria are non‑negotiable for production signoff? Options: Sustained throughput KPI, Worst‑case latency bound, ASIL evidence package, Thermal envelope compliance, Supply commitment, All of the above
• What quantitative KPIs and test scenarios should we include in a joint verification checklist? Options: Per‑network FPS/latency, End‑to‑end detection latency, False positive/negative rates, Power vs performance curves, Scenario pass/fail matrix
• What would success look and feel like to your team six months after integration?
• Who needs to be on the scoreboard for success to be celebrated internally (teams, roles)? Options: VP/Director of ADAS, Perception Leads, Functional Safety, Hardware Architecture, Procurement, Other

What Would It Take to Move Forward (Smallest, Least‑Risky Step)

• What is the smallest pilot or evidence package that would materially accelerate your decision? Options: Reference SoC evaluation board, Head‑to‑head benchmark on our nets, Joint scenario workshop, Safety evidence draft, Supply sketch/letter of intent
• What access do you need to our toolchain, CI, or benchmarks to validate your networks quickly? Options: Full SDK access, Sandboxed compiler + docs, Benchmark scripts and sample data, Remote assisted tests, Not sure — need a proposal
• What contractual or risk‑mitigating terms would make a pilot acceptable (e.g., limited quantity, milestone refunds, acceptance gates)?
• Realistically, when could you commit engineering time to a joint evaluation sprint? Options: Immediately, In 1–4 weeks, 1–3 months, Longer / need planning
• Who should we schedule a technical kickoff with, and what format helps you best (workshop, hands‑on lab, or remote review)? Options: Onsite workshop, Remote workshop, Hands‑on lab with engineering support, Asynchronous data exchange + review

Tell Us About the Program You're Betting On

• Please share the program name, vehicle segment, and planned Start‑of‑Production (SOP) date.
• What is the primary capability you need this compute platform to deliver for the program (e.g., L2+ highway pilot, L3 urban chauffeur, sensor-fusion backbone)? Options: L2+/Assisted driving, L3 Conditional Automation, L4 Pilot/Low-speed autonomy, ADAS feature acceleration, Other
• Who will be the core decision-makers for silicon selection on this program? Select all who apply and name the owner where possible. Options: OEM engineering (platform), OEM safety/validation, Tier‑1 ADAS HW lead, Tier‑1 perception lead, Procurement, Other
• How stable are your high‑level requirements today—are they still evolving, mostly set, or locked in? Options: Evolving significantly, Mostly set with some open items, Locked in for architecture selection, Unsure
• What would you consider a clear early signal that this program is on track (one measurable success in the next 3–6 months)?

If Safety Were the Organizing Principle, What Would Change?

• If you had to tighten safety targets tomorrow, which functions would you immediately move toward ASIL‑D?
• What is your target ASIL per system or function today? Options: ASIL-A, ASIL-B, ASIL-C, ASIL-D, Not yet defined
• How are you currently partitioning safety responsibilities between the SoC, middleware, and application layers?
• Which specific safety mechanisms do you expect from a silicon partner (e.g., lockstep cores, ECC memory, safety island, safety-certified RTOS, traceability artifacts)? Options: Lockstep/dual cores, ECC and parity, Safety island/hardware supervisor, Safety-certified RTOS, Deterministic watchdogs, Traceable build artifacts, Other
• How would failing to achieve your safety evidence milestones affect OEM signoff or program timing? Options: Minor delay with mitigation, Major program delay, OEM rejection risk, Unsure

Are You Designing for Peak Performance or Predictable Behavior?

• What are your primary inference throughput requirements (per sensor and aggregated) in TOPS or inferences/sec?
• What worst‑case end‑to‑end latency target do you require on the safety‑critical perception → decision path? Options: <1 ms, 1–5 ms, 5–10 ms, 10–25 ms, 25–50 ms, >50 ms
• Which network families and approximate model sizes do you plan to run in production (e.g., ResNet50 25M params, EfficientDet, PointPillars, transformer backbones)?
• Which optimization strategies are you committed to or planning to use for production (select all that apply)? Options: INT8 quantization, Pruning, Knowledge distillation, Dynamic batching, Custom kernels, None yet / undecided
• How often do you expect model architecture or topology changes during development and post‑launch? Options: Continuous/weekly, Monthly, Quarterly, Rare/annual, Unsure

Where Will Heat and Power Become Your Uninvited Guests?

• Given your planned integration location, what sustained thermal/power budget is available for the compute module? Options: <5 W, 5–10 W, 10–25 W, 25–50 W, 50–100 W, >100 W
• How much of that budget can be allocated to peak bursts vs sustained operation—please quantify if possible.
• What cooling strategies are feasible in the vehicle architecture (select all realistic options)? Options: Passive conduction, Small forced‑air, Liquid cold plate, Thermal interface to chassis, Dedicated fan/duct, Not decided
• Which sensor placements or enclosure constraints limit thermal dissipation and where are they on the vehicle?
• How much headroom do you want reserved for future algorithm growth or model complexity increases? Options: 0–10%, 10–25%, 25–50%, >50%

What Happens When the Model Changes—Do You Break the Car?

• When you deploy a new model, what level of bit‑for‑bit determinism or reproducibility is required across builds and platforms? Options: Bit‑exact required, Functionally equivalent acceptable, Selective determinism by use case, Undecided
• How do you plan to validate each model update against the vehicle safety case (e.g., shadow‑mode fleet, HIL, regression suites)? Options: Shadow/observer fleet, HIL testing, Extended lab regression, Field canary rollout, Other
• Which verification artifacts do you need from a silicon supplier to accept a model update (profiling logs, cycle‑accurate traces, timing budgets, failure-mode repro)? Options: Profiling traces, Cycle‑accurate simulation, Latency distributions, Power traces, Error/fault injection reports, Other
• How many labeled verification datasets do you have for safety‑critical scenarios and how representative are they of edge cases? Options: Extensive and representative, Moderate coverage, Sparse / limited edge cases, None available
• If validation shows a performance regression after a model update, what rollback or mitigation path do you require? Options: Immediate rollback, Gradual canary rollback, Activate fallback safety function, Investigate before rollback, Undecided

Who Really Owns Risk When Things Go Wrong?

• How are acceptances and liabilities split today between Tier‑1, OEM, and silicon supplier for software and hardware safety failures?
• Which safety artifacts do you already have or expect to produce (HARA, FMEDA, safety plan, safety case, verification matrix)? Options: HARA, FMEDA, Safety plan, Safety case, Verification matrix, None / starting
• What acceptance criteria will OEM demand for safety and performance prior to production signoff?
• Describe your preferred joint‑issue resolution workflow when cross‑vendor root cause analysis is required.
• How tolerant is the program to delays in safety artifact delivery (e.g., months of delay)—and how long have you been carrying that tolerance? Options: Very tolerant (months), Moderately tolerant (weeks), Low tolerance (days), Not tolerant / showstopper

If Supply Tightness Were a Showstopper, What's Your Plan B?

• What supply constraints matter most for you: AEC‑grade components, single‑source dies, long lead passive parts, or certified packaging? Options: AEC‑grade parts, Single‑source dies, Long‑lead passives, Qualified packaging, None critical
• Do you require guaranteed wafer/assembly capacity commitments or volume reservations from a silicon supplier? Options: Full capacity reservation, Partial reservation, No reservation required, Undecided
• How would a 6–12 month slip in silicon supply affect your program milestones and OEM commitments? Options: Manageable with mitigation, Requires reprioritization, Major program risk/delay, Program would be cancelled
• Which certifications or OEM‑specific compliance items beyond ISO 26262 and AEC‑Q100 are mandatory for your program?
• What contingency plans (alternate suppliers, software retargets, feature de‑scoping) are acceptable if supply tightness emerges? Options: Alternate silicon vendor, Software optimization to lower power, Feature de‑scope, Late binding to other modules, Unsure

Cross‑Functional Truths: Who Will Fight for This Choice?

• Which functions will have veto power or final signoff on the silicon decision (select all that apply)? Options: OEM platform engineering, OEM safety, Tier‑1 ADAS engineering, Product management, Procurement, Other
• Who will act as the program's single technical point of contact between your teams and our engineering organization?
• Where do you anticipate the hardest cross‑functional tradeoffs will be (performance vs thermal, cost vs reliability, time vs evidence)?
• How much influence does the OEM have over the Tier‑1's final supplier choice on this program? Options: OEM controls choice, OEM heavily influences, OEM consulted only, OEM rarely intervenes
• What internal metrics or KPIs will your stakeholders use to evaluate whether this silicon selection was the right one after 12 months in program?

If We Could Remove One Uncertainty Today, Which Would Change Everything?

• Select the top three risks that, if removed, would materially accelerate your decision and why (pick up to three). Options: Safety evidence gap, Thermal/power uncertainty, Latency/jitter risk, Model performance unknowns, Supply/capacity risk, Software toolchain maturity
• What single piece of evidence or data from a silicon partner would most reduce your program risk (e.g., representative workload benchmark, in‑vehicle thermal trace, third‑party safety audit)?
• How quickly would you need that evidence for it to meaningfully change your schedule? Options: Within 2 weeks, 1 month, 1–3 months, 3–6 months, >6 months
• Would you be open to a focused pilot that validates one safety‑critical use case on our platform before a broader evaluation? Options: Yes, immediately, Yes, with scoped timeline, Maybe, need more details, No
• If we propose next steps (pilot scope, artifacts, timeline), who should own the joint plan on your side and when can we schedule a follow‑up to agree scope?

Quick Snapshot: Where We Begin

• In one sentence, what is this ADAS/autonomy program trying to deliver and what is your target SOP?
• Who is the immediate customer for this program? Options: OEM (vehicle OEM), Tier‑1 supplier, Mobility service provider, System integrator, Other
• How would you describe the program timeline right now? Options: Within 12 months, 12–24 months, 2–3 years, 3–5 years, 5+ years, Unsure
• What single metric will be used to declare this program a technical success?
• Who on your side will be the core decision-makers and approvers for compute selection? Options: VP Engineering, Director ADAS/Autonomy, Functional Safety Lead, HW Architecture Lead, SW/Perception Lead, Procurement, OEM Approval Authority, Other
• What existing commitments (contracts, supply agreements, design freezes) should we be aware of?

Are You Settling for 'Good Enough' Compute?

• How many times this year have you had to slow, simplify, or redesign an algorithm because the SoC couldn't meet requirements? Options: Never, 1–2 times, 3–5 times, 6–10 times, More than 10 times
• Describe the most recent compute bottleneck you encountered and the real program impact it caused.
• Which workloads are most constrained on your current compute platform? Options: Object detection, Semantic segmentation, Tracking/fusion, End-to-end planning, Path planning/control, Model training/finetune, Other
• What are the latency targets for the safety‑critical inference path at the percentile you care about (e.g., P99)? Options: ≤5 ms, 5–10 ms, 10–20 ms, 20–50 ms, >50 ms, Unsure
• What is the per‑SoC thermal/power envelope you must stay within? Options: <5 W, 5–10 W, 10–20 W, 20–40 W, >40 W, Unsure
• How would missing your inference‑per‑watt target change your architecture or product decisions?
• Which model families and input resolutions are you targeting for production (briefly list examples)?

Who Really Calls the Shots?

• If your top technical recommendation were overridden by cost or schedule, what would that do to safety, performance, or program risk?
• Which organizations influence or own the silicon decision in this program? Options: OEM Architecture Team, OEM Procurement, Tier‑1 Product Management, Tier‑1 HW Architecture, Functional Safety Team, SW/Perception Leads, Systems Integration, Legal/Regulatory, Other
• Which single party holds final approval authority for compute selection? Options: OEM, Tier‑1, Joint approval, Program steering committee, Unsure
• What procurement or compliance constraints matter most (e.g., single‑source, QPL, localization, ITAR)?
• How regularly do approval gates slip on your programs, and what are the usual root causes? Options: Almost never, Occasionally (1–2 gates), Often (multiple gates), Frequently derailed, Unsure
• Who on the OEM side would we need to brief to accelerate technical approval, and what evidence do they prioritize?

What Keeps Your Engineers Up at Night?

• When was the last time an algorithm passed lab tests but failed in a vehicle; what failed and why did it surprise you?
• Which verification gaps cause the most anxiety: determinism, toolchain maturity, data representativeness, or safety traceability? Options: Determinism/repeatability, Toolchain immaturity, Insufficient representative data, Incomplete safety artifacts, Sensor calibration variability, Other
• What test benches and vehicle access do you currently have for validation? Options: Full vehicle access (fleet), Limited vehicle access, HiL only, Simulation/virtual only, None
• How accessible are representative datasets and logs for partners to benchmark against your workloads? Options: Openly shareable, Shareable under NDA, Limited sample share only, Not shareable (IP/contractual), Unsure
• How often do intermittent latency spikes or tail‑events appear in your field logs? Options: Never, Rarely, Occasionally, Frequently, Constantly
• When integration stalls occur, what’s the typical time to resolution and what usually causes the delay?
• Which emotional reactions do integration pain points usually trigger in your teams (e.g., frustration, fear of rework, resignation)? Options: Frustration, Urgency/stress, Fear of rework, Loss of confidence, Motivation to innovate, Other

If You Could Wave a Wand…

• If one non‑negotiable requirement could be met by a compute partner, what would it be?
• Choose the top three technical priorities you would insist on in an ideal platform. Options: Maximum inference throughput/watt, Deterministic low latency, ASILD functional safety support, Long‑term software lifecycle support, OTA/update friendliness, Firm supply commitments, Mature toolchain & ecosystem, Other
• What would a complete and convincing verification package look like to you?
• How would you like model updates, retraining, and deployment to be handled across the vehicle lifetime? Options: Tier‑1 manages OTA updates, OEM manages OTA updates, Joint process with shared tooling, Manual update via service updates, Other
• What commercial or supply guarantees would make you feel secure enough to commit (e.g., fixed volumes, price protection, priority allocation)?
• If we could deliver one early technical win in the first 90 days, what should it be to build trust?

What Would Block a Smooth Production Handover?

• Imagine a three‑week production‑signoff delay — which missing artifact is most likely to cause it?
• Do you currently have CI and toolchain access provisioned for partners and system integrators? Options: Full access for partners, Partial access under NDA, Internal only, No CI/toolchain access yet
• Which verification artifacts must be completed and owned at handover? Options: Performance baselines/benchmarks, Safety case documentation, Traceability matrices, Testbench configurations, Regression suites, Calibration datasets, Other
• Who on your team will own and maintain verification artifacts after handover (name roles)?
• How confident are you in your dataset handshakes (licenses, NDAs, access mechanisms) with sensor and algorithm suppliers? Options: Very confident, Somewhat confident, Not confident, No handshake exists yet
• What logistics or vehicle‑access constraints could realistically slow integration sprints or late validation?

How Will We Prove It Works?

• If the OEM asks for a demo they can reproduce in their labs, what would you struggle to reproduce today?
• Which acceptance metrics will the OEM insist on for production readiness? Options: Inference throughput, P99 (or chosen percentile) latency, Power/Watt under load, Sustained temperature under load, Functional safety traceability, Perception accuracy (mAP), Other
• Which testbeds should we jointly provision to accelerate validation (pick all that apply)? Options: Vehicle‑in‑the‑loop (VIL), HiL rigs, Edge sensor integration lab, Cloud benchmarking cluster, Silicon evaluation boards, Automated regression farm
• What would a clear pass/fail criterion look like for OEM signoff on performance and safety?
• Who will sign the OEM acceptance paperwork and what is the expected signoff timeline?
• Would co‑authoring test protocols and acceptance criteria reduce risk for you? Options: Yes, Maybe, No

How Do We Want to Work Together?

• If this partnership were to fail, what single reason would you point to first—tech mismatch, missed schedules, unclear responsibilities, or commercial issues? Options: Tech mismatch, Missed schedules, Unclear responsibilities, Commercial terms, Other
• Which collaboration model would you prefer for the next 12 months? Options: Tight co‑engineering with weekly sprints, Milestone‑based checkpoints, Advisory with occasional deep dives, Toolchain and training centric, Other
• Who should be the named owners on both sides for integration, safety case delivery, and supply commitments?
• What cadence of status reviews would feel most effective to avoid late surprises? Options: Weekly, Bi‑weekly, Monthly, Milestone‑driven, Ad hoc as needed
• What escalation path would you like for high‑severity blockers (who, how, and expected response SLAs)?
• What early deliverable or milestone would make you feel confident we’re the right compute partner?

Who Are We Bringing Into This Room?

• Please tell us your role and the primary responsibilities you own for this ADAS/autonomy program (pick the closest match). Options: VP / Head of ADAS or Autonomy, Director / Manager, Autonomous Driving Platform, Lead Systems Architect, Perception / CV Lead, Functional Safety Engineer, HW Architecture / SoC Lead, Procurement / Program Manager, Other (please specify)
• Which organization are you representing on this program? Options: OEM - Vehicle OEM, Tier‑1 Supplier, Tier‑2 Supplier, Systems Integrator, Other
• What is the current program phase (early concept, requirements, design, integration, validation, production)? Options: Concept / Feasibility, Requirements definition, Architecture design, Integration / bring‑up, Validation / verification, Production ramp
• What is your target start‑of‑production (SOP) year? Options: Within 12 months, 1–2 years, 2–3 years, 3–5 years, 5+ years / exploratory
• Who is the OEM-level decision owner for silicon/platform selection on this program?

What If This Choice Really Matters — Are We Ready for That?

• If the chosen SoC under‑delivers in production, what would that mean for your program—schedule slips, safety risk, cost overruns, or program cancellation? Options: Schedule slips, Reduced feature set, Increased cost, Safety/regulatory risk, OEM rejection of supplier choice, Other / unsure
• Tell us about a past platform decision that caused pain—what went wrong, and what were the consequences?
• How much margin do you currently have between your model/latency needs and the platform you are using or targeting? Options: Wide margin (>40%), Healthy margin (20–40%), Tight margin (5–20%), No margin / over budget (0–5%), Unknown
• What would you most worry about if a new silicon supplier could not meet a critical milestone? Options: Late HW bring‑up, Unstable drivers / SW, Thermal failures in vehicle, Insufficient safety evidence, Supply shortage, Other
• How confident are you in the supplier’s ability to provide long‑term software support and lifecycle updates? Options: Very confident, Confident, Somewhat confident, Not confident, Unsure

What’s Actually Burning on the Bench Right Now?

• Where are your current compute bottlenecks—throughput, worst‑case latency, thermal limits, memory bandwidth, or software determinism? Options: Throughput (infer/sec), Worst‑case latency / tail latency, Thermal / power budget, Memory bandwidth, I/O / sensor fusion bottleneck, Software determinism / jitter, Other
• Which perception or planning networks are causing the most deployment friction (e.g., detection, segmentation, tracking, planner)? List model names or topologies if possible.
• How often do you hit safety‑path deadlines (misses per 1000 runs or percent of scenarios with failed timing)? Options: Never, Rarely (<1%), Occasionally (1–5%), Often (5–20%), Frequently (>20%), Don't measure / unknown
• When a compute limit is reached, what do you typically sacrifice—model fidelity, sensor rate, redundancy, or function scope? Options: Model fidelity (smaller nets), Lower sensor frame rate, Remove redundancy, Reduce functionality (graceful degradation), Other
• How does this bottleneck feel to your team—annoying but tolerable, a career risk, or company‑level anxiety? Options: Annoying / tolerable, Creates project delays, Serious safety/certification concern, Career / leadership pressure, Other

Who Actually Holds the Keys (and What Evidence Will Convince Them)?

• Which stakeholders must sign off on silicon selection and what does each care about most (e.g., OEM: reliability, Safety Eng: ASIL traceability, Procurement: cost/supply)?
• Which party has final approval authority at the OEM level for this supplier choice? Options: OEM central procurement, OEM ADAS engineering, OEM safety certification board, Tier‑1 prime integrator, Shared decision
• What types of evidence will your OEM/safety team require to accept the platform (bench performance, HW safety docs, ISO 26262 artifacts, 3rd‑party audits)? Options: Performance benchmarks vs target models, ASIL evidence / safety case artifacts, Thermal and environmental testing, Long‑term supply commitment, Software toolchain verification, Reference vehicle integration
• How do you prefer evidence to be presented—raw data and logs, curated test narratives, independent lab reports, or interactive demos? Options: Raw data & logs, Curated test narratives, Independent lab report, Interactive scenario demos, All of the above
• How long does your approval cycle typically take once complete evidence is submitted? Options: <2 weeks, 2–6 weeks, 6–12 weeks, 3–6 months, Longer / varies widely

If Performance Is King, What Does the Throne Look Like?

• What are your target inference‑per‑watt and worst‑case latency numbers for the safety‑critical stack (please provide numeric targets where possible)?
• Which ASIL level(s) must the platform support for critical functions on this program? Options: ASIL A, ASIL B, ASIL C, ASIL D, Multiple levels across functions, Unsure / evaluating
• What thermal and power budgets do you have per compute module (W) and are there moments with temporary power spikes we should plan for?
• Which sensor configurations are in play (camera only, camera+radar, lidar fusion, high‑channel radar), and which must be supported in production? Options: Camera only, Camera + Radar, Camera + Lidar, Full sensor fusion (Camera+Radar+Lidar+Ultrasonic), Other / custom sensor stack
• What model formats, runtimes, or quantization strategies must be supported (e.g., FP16, INT8, ONNX, TensorRT, custom runtimes)? Options: FP32/FP16, INT8, ONNX, TensorRT / equivalent, Custom runtime, Unknown / flexible
• How important is software determinism (bounded latency) vs average throughput for your safety case? Options: Determinism is critical, Both equally important, Throughput is priority, Depends on function

What Would It Take to Trust 'Production‑Ready'?

• What acceptance criteria will you use to declare the platform production‑ready (performance thresholds, ASIL evidence, environmental tests, supply agreements)?
• Do you have vehicle access, test benches, and datasets available for joint validation, or will you rely on supplier labs? Options: We have vehicle/test bench access, We have datasets but limited vehicle access, We will rely on supplier labs, Mixed / partial access
• Which verification artifacts must have a defined owner on both sides (e.g., CI pipelines, regression suites, safety traceability matrices)? Options: CI pipelines, Regression suites, Safety trace matrices, Verification test plans, Artifact ownership is not yet defined
• How do you prefer to run joint validation—periodic gated milestones, continuous CI integration, or an intensive gated lab validation window? Options: Gated milestones, Continuous CI, Intensive lab window, Combination
• What turnaround time do you need for bug fixes or SW updates during integration to keep your schedule healthy? Options: 24–72 hours, 1–2 weeks, 2–4 weeks, Depends on severity, Unsure
• Who on your team will be the primary contact for verification handshakes and SLA escalation?

Supply, Contracts, and the Path to Production — Are We Built for the Long Run?

• What are your forecasted unit volumes for initial production and year‑3 ramp? Options: Prototype / low volume (<1k), Low volume (1k–10k), Medium (10k–100k), High (>100k), Unsure
• How risky would single‑sourcing this SoC be for your program, and do you have qualified fallbacks? Options: Low risk — fallbacks ready, Moderate risk — partial fallbacks, High risk — no fallbacks, Undecided
• Which commercial commitments are non‑negotiable (supply lead times, long‑term pricing, warranty, IP/license terms)?
• Have you experienced supplier shortages or allocation issues that impacted your program before? Options: Yes — significant impact, Yes — minor impact, No, Not applicable
• What cadence do you need for production readiness reviews and supply checkpoints with your vendor partners? Options: Weekly, Bi‑weekly, Monthly, Quarterly, Event‑driven
• What would make you feel comfortable about a supplier's production commitments (e.g., dual fabs, guaranteed MOQ, escrowed IP)?

How Will We Know This Was the Right Choice?

• What are the top 3 success signals you will look for in the first 6–12 months after SOP?
• Which metrics should we report regularly to prove performance and safety in the field (e.g., sustained inference throughput, in‑vehicle latency percentiles, field failure rate)? Options: Sustained inference throughput, 99th/99.9th percentile latency, Field failure rate / incidents, Thermal throttle events, OTA update success rate, Other
• How would you like learnings and lifecycle issues to be communicated—shared channel, monthly reviews, joint Jira, or direct escalation? Options: Shared channel (Slack/MS Teams), Monthly review, Joint issue tracker (Jira), Direct escalation path
• If early field data shows divergence from lab results, what is your preferred remediation path? Options: Immediate rollback, Patch / OTA, Feature throttling, Joint hotfix task force, Other
• Who will own lessons‑learned and continuous improvement for software and safety artifacts after production? Options: OEM engineering, Tier‑1 integration team, Supplier (SoC vendor), Joint governance board

Next Steps — What Would Make This Conversation Immediately Valuable?

• Which immediate deliverable would be most helpful right now: a tailored performance estimate for your models, a joint risk register, a proposed verification plan, or a supply commitment summary? Options: Performance estimate for our models, Joint risk register, Proposed verification plan, Supply commitment summary, All of the above
• Who should we include from your side in a technical kickoff to avoid rework (names, roles, contact emails if available)?
• What timing works for an initial joint workshop to map models to scenarios and run an early performance‑for‑power analysis? Options: This week, Next 2 weeks, Within a month, Next quarter, Undecided
• Are there any non‑technical constraints we should know about up front (export controls, preferred vendors, contractual terms)?
• Is there anything we haven’t asked that would change how we prioritize support for your program?