The first commercial STELLAR node. Higher compute density, scheduled customer service windows, automated delivery with receipts. Demo workloads become paid workloads.
Move from engineering demonstration to early commercial workloads with a stronger compute module and repeatable service operations.
Node-1 proves the foundation. Node-2 turns it into a product. Compute density increases through a denser FPGA + GPU coprocessor configuration; storage capacity steps up to support customer datasets; scheduled service windows replace demo executions; and delivery is automated end-to-end so STELLAR can take a customer job at 09:00 and return a receipt-bearing result before the next pass.
Node-2 introduces the first formal customer service contract: scheduled execution windows, priority classes, delivery deadlines, integrity guarantees, and replayable evidence. The same operations console and customer API used since GroundLab now carry SLA semantics. Customers can subscribe to a window class instead of asking for a one-off run.
A scheduled service window only holds if power, thermal, and contact stay inside their envelopes. Node-2’s scheduler models all three jointly solar generation, eclipse, battery state, radiator hot-case, and pass timing and offers windows the spacecraft can actually keep. When margins narrow, lower-priority classes degrade gracefully instead of breaching SLAs.
The profile is the mission’s physical truth every architectural choice descends from these numbers.
The mission is the sum of these subsystems and the way they constrain each other.
| System | Specification | Design note |
|---|---|---|
| Compute fabric | FPGA + radiation-tolerant GPU coprocessor (~120 TOPS) | Adds a coprocessor lane for higher-throughput AI inference and larger-batch image processing while preserving the deterministic FPGA path for autonomy workloads. |
| Management CPU | Quad-redundant GR740 cluster | Workload scheduler, customer-API gateway, FDIR, telemetry mux. Dual-string software with cross-channel checkpointing. |
| Storage path | 8 TB hot pool + 32 TB persistent pool + 256 GB priority queue | Customer datasets persist across passes. Priority queue ranks delivery against contact windows, customer SLAs, and mission-safe thermal constraints. |
| Thermal | Larger deployable radiators + active loop control | Active thermal control adjusts loop flow against compute load profile. Hot-case capacity sized for sustained 1.4 kW rejection at end-of-life. |
| Comms | X-band downlink + dual S-band TT&C, multi-site ground | Multi-pass delivery across two or three ground sites narrows the gap between job submission and customer receipt to under 4 hours typical. |
| Service contract | SLA-backed scheduling, priority classes, replayable receipts | API contract gains scheduling primitives (window-class subscription, deadline guarantees) and integrity primitives (signed receipts, replay log). |
Each mission is justified by a specific new capability not just a larger payload.
Scheduled customer workload execution windows
Persistent orbital storage with priority downlink rules
Automated job receipt, execution, and delivery workflow
Higher thermal and power envelope for more useful compute density
Multi-pass delivery reconciliation across multiple ground sites
Service-level priority classes with graceful degradation
The plan is broken into reviewable phases. Each phase ends in a deliverable a stakeholder can read.
Scheduling primitives, SLA classes, evidence guarantees specified with three to five design-partner customers.
Compute coprocessor integration, scheduler architecture, and service-contract review.
Full vibration / TVAC / TID qualification of the coprocessor + thermal stack.
Commissioning campaign scheduler validation, ground-cloud SLA loop, first paid customer window.
Steady-state commercial operations. Telemetry feeds Node-3 cluster planning.
A serious mission tracks risks honestly. Closed, mitigated, in design, and open items are listed without softening.
| Risk | State | Description |
|---|---|---|
| GPU coprocessor SEU rate at design altitude | In design | Total Ionizing Dose and Single-Event Upset rates at 500–550 km mid-inclination drive coprocessor selection and shielding strategy. |
| SLA breach under thermal hot case | In design | Scheduler must demote priority classes before breaching SLAs. Validated through orbit-condition replay against the GroundLab twin. |
| Multi-site ground reconciliation | Mitigated | Receipts are signed and replay-safe; ground-cloud reconciliation closes when all-of-N signatures present. Tested in Node-1 telemetry. |
| Customer concentration risk | Open | Node-2 is sold to a small number of design partners. Diversification is a Node-3 commercial gate, not Node-2. |
The right way to read a mission page: claim → evidence → state. If a row is in the wrong state, the page is what is broken not the program.
| Claim | Evidence | State |
|---|---|---|
| Scheduler delivers contracted SLA classes under nominal thermal/power state. | On-orbit scheduler logs reconciled against GroundLab predictions | Planned |
| Coprocessor maintains predictable throughput under design SEU rate. | TID + heavy-ion radiation testing + on-orbit performance log | Planned |
| Customer job → result median latency under 4 hours. | Multi-site ground reconciliation telemetry, 90-day campaign | Planned |
| Hot-case radiator capacity sustains 1.4 kW rejection at EOL. | Thermal correlation + on-orbit telemetry | Draft |
Each mission is a step on a single staircase. These bridges spell out what comes from the prior step and what becomes possible after.
Node-2 inherits the Node-1 hosted-payload architecture and replaces the host bus with a dedicated module same compute lineage, larger envelope, scheduler upgraded for SLA-backed service.
← Open Node-1Once Node-2 holds SLAs in commercial operation, Node-3 introduces a second coordinated node and proves clustering: distribution, replication, and failover across multiple data-center assets.
Open Node-3 →