# Athena Engine Board v2: NUC Form Factor

## Design Philosophy

Compact NUC-sized unit (like Mac Mini) containing both CPU/GPU and NPU domains. Multiple units connect via PCIe/PCIe lanes for distributed computing. Each unit is self-contained; stacking adds capacity.

---

## Form Factor

**Dimensions:** 170mm x 170mm x 40mm (Mac Mini: 197mm x 197mm x 36mm)

**Enclosure:** CNC aluminum, passive cooling, unified thermal design

```
Top View:
┌─────────────────────────────────────────┐
│                                         │
│    ┌─────────────────────────────┐     │
│    │      Athena NUC             │     │
│    │      (170mm x 170mm)        │     │
│    └─────────────────────────────┘     │
│                                         │
│    ┌──────────┐  ┌──────────────────┐  │
│    │ CPU/GPU  │  │ NPU + Graph SSD  │  │
│    │ + System │  │ (isolated)       │  │
│    │ SSD      │  │                  │  │
│    └──────────┘  └──────────────────┘  │
│                                         │
│    ┌─────────────────────────────┐     │
│    │      Power + Cooling        │     │
│    └─────────────────────────────┘     │
│                                         │
└─────────────────────────────────────────┘

Side View:
┌─────────────────────────────────────────┐
│ 40mm                                   │
│ ┌─────────────────────────────────────┐│
│ │ Top: Thermal plate (aluminum)       ││
│ ├─────────────────────────────────────┤│
│ │ CPU/GPU + RAM + System SSD          ││
│ ├─────────────────────────────────────┤│
│ │ Bridge Controller                   ││
│ ├─────────────────────────────────────┤│
│ │ NPU + NPU RAM + Graph SSD           ││
│ ├─────────────────────────────────────┤│
│ │ Bottom: Power + Connectors          ││
│ └─────────────────────────────────────┘│
└─────────────────────────────────────────┘
```

---

## Internal Layout

```
┌─────────────────────────────────────────────────────────────────┐
│                    Athena NUC (Top View)                        │
│                                                                 │
│  ┌───────────────────────────────────────────────────────────┐ │
│  │                    THERMAL PLATE                          │ │
│  │                 (Aluminum, passive)                        │ │
│  └───────────────────────────────────────────────────────────┘ │
│                                                                 │
│  ┌─────────────────────┐  ┌─────────────────────────────────┐ │
│  │   CPU/GPU Domain    │  │       NPU Domain               │ │
│  │                     │  │                                 │ │
│  │  ┌───────────────┐  │  │  ┌───────────────┐             │ │
│  │  │ CPU (ARM/x86) │  │  │  │ NPU           │             │ │
│  │  │ + GPU         │  │  │  │ (Hailo-8L)    │             │ │
│  │  └───────────────┘  │  │  └───────────────┘             │ │
│  │                     │  │                                 │ │
│  │  ┌───────────────┐  │  │  ┌───────────────┐             │ │
│  │  │ CPU RAM       │  │  │  │ NPU RAM       │             │ │
│  │  │ (8-16GB)      │  │  │  │ (8GB)         │             │ │
│  │  └───────────────┘  │  │  └───────────────┘             │ │
│  │                     │  │                                 │ │
│  │  ┌───────────────┐  │  │  ┌───────────────┐             │ │
│  │  │ System SSD    │  │  │  │ Graph SSD     │             │ │
│  │  │ (1-2TB)       │  │  │  │ (2-4TB)       │             │ │
│  │  └───────────────┘  │  │  └───────────────┘             │ │
│  │                     │  │                                 │ │
│  │  ┌───────────────┐  │  │                                 │ │
│  │  │ rPI Slot      │  │  │                                 │ │
│  │  │ (optional)    │  │  │                                 │ │
│  │  └───────────────┘  │  │                                 │ │
│  │                     │  │                                 │ │
│  └──────────┬──────────┘  └──────────────┬─────────────────┘ │
│             │                            │                     │
│             │    ┌──────────────────┐    │                     │
│             └───►│ Bridge Controller│◄───┘                     │
│                  │ (Isolation)      │                          │
│                  └──────────────────┘                          │
│                                                                 │
│  ┌───────────────────────────────────────────────────────────┐ │
│  │                 REAR CONNECTORS                            │ │
│  │  [Power] [Ethernet] [USB] [PCIe/PCIe] [HDMI] [GPIO]     │ │
│  └───────────────────────────────────────────────────────────┘ │
│                                                                 │
└─────────────────────────────────────────────────────────────────┘
```

---

## Component Specifications

### CPU/GPU Domain

**Options:**

| Option | CPU | GPU | RAM | Power | Use Case |
|--------|-----|-----|-----|-------|----------|
| **A (ARM)** | BCM2712 (4xA76) | Mali-G610 | 8GB LPDDR4X | 10W | Edge, low-power |
| **B (ARM Pro)** | RK3588 (4xA76+4xA55) | Mali-G610 | 16GB LPDDR4X | 15W | Balanced |
| **C (x86)** | Ryzen 5 7600U | Radeon 680M | 16GB DDR5 | 28W | Performance |
| **D (x86+GPU)** | Ryzen 7 7840U | Radeon 780M | 32GB DDR5 | 40W | Workstation |

**System NVMe SSD:**
- 1-2TB NVMe PCIe 3.0 x4
- For OS, applications, user data

**rPI Slot:**
- Accepts Raspberry Pi 5 (or 4)
- PCIe 3.0 x1 to CPU
- USB 3.0 for power + data
- GPIO passthrough
- Optional: dedicated I/O controller

### NPU Domain (Isolated)

**NPU:**
- Hailo-8L (26 TOPS, 2.5W) - Base
- Hailo-8 (130 TOPS, 5W) - Upgrade
- Future: Custom Athena NPU

**NPU RAM:**
- 8GB LPDDR5 (dedicated, not shared)

**NPU Graph NVMe SSD:**
- 2-4TB NVMe PCIe 3.0 x4
- **Directly connected to NPU only**
- CPU cannot access this drive
- Models + Knowledge Graph + Embeddings

### Bridge Controller

**Purpose:** Isolate CPU/GPU from NPU; enable defined communication

**Hardware:** Lattice ECP5 FPGA (or ASIC in production)

**Functions:**
- Message queue (4KB messages)
- DMA engine (bulk transfer)
- Hardware-enforced isolation
- Protocol translation (AIDP)

**Bandwidth:** 10 GB/s (PCIe 3.0 x4 equivalent)
**Latency:** <10μs message, <100μs DMA

---

## Rear Connectors

```
┌─────────────────────────────────────────────────────────────────┐
│                    REAR PANEL                                   │
│                                                                 │
│  ┌──────┐ ┌──────┐ ┌──────┐ ┌──────┐ ┌──────┐ ┌──────┐      │
│  │Power │ │ Eth  │ │ USB  │ │PCIe/ │ │ HDMI │ │ GPIO │      │
│  │ USB-C│ │ GbE  │ │3.2x2 │ │PCIe  │ │ 2.0  │ │40-pin│      │
│  │      │ │ PoE  │ │      │ │ x8   │ │      │ │      │      │
│  └──────┘ └──────┘ └──────┘ └──────┘ └──────┘ └──────┘      │
│                                                                 │
│  ┌──────┐ ┌──────┐                                             │
│  │ USB  │ │ JTAG │                                             │
│  │ 2.0  │ │ Debug│                                             │
│  └──────┘ └──────┘                                             │
│                                                                 │
└─────────────────────────────────────────────────────────────────┘
```

**Connector Details:**

| Connector | Specification | Purpose |
|-----------|---------------|---------|
| Power | USB-C PD (100W) | Main power input |
| Ethernet | GbE with PoE | Network, power (optional) |
| USB 3.2 | 2x Type-A (10Gbps) | External storage, peripherals |
| PCIe/PCIe | x8 (16 GB/s) | Stack to other Athena units |
| HDMI | 2.0 (4K60) | Display output |
| GPIO | 40-pin header | Sensors, OSE machines, debug |
| USB 2.0 | 1x Type-A | Low-speed devices |
| JTAG | 10-pin | Hardware debug |

---

## Stacking / Distribution Model

### Single Unit

```
┌─────────────────────────────────────┐
│         Athena NUC                  │
│  (CPU/GPU + NPU + Storage)         │
└─────────────────────────────────────┘
```

### Dual Stack (CPU + NPU Expansion)

```
┌─────────────────────────────────────┐
│         Athena NUC #2               │
│  (NPU + Graph SSD expansion)       │◄─── PCIe/PCIe x8
├─────────────────────────────────────┤
│         Athena NUC #1               │
│  (CPU/GPU + NPU + Storage)         │
└─────────────────────────────────────┘
```

### Quad Rack (Distributed)

```
┌─────────────────────────────────────┐
│         Athena NUC #4               │
├─────────────────────────────────────┤
│         Athena NUC #3               │
├─────────────────────────────────────┤
│         Athena NUC #2               │
├─────────────────────────────────────┤
│         Athena NUC #1               │
└─────────────────────────────────────┘
         │
         ▼
    Ethernet to network
```

### Expansion via PCIe/PCIe

**Stacking Connector (rear):**
- PCIe 3.0 x8 (16 GB/s bidirectional)
- Power pass-through (12V/10A)
- I2C control bus

**Expansion Modules:**

| Module | Purpose | Interface |
|--------|---------|-----------|
| NPU Expansion | Additional NPU + Graph SSD | PCIe/PCIe x8 |
| GPU Expansion | Discrete GPU (RTX 4060) | PCIe/PCIe x8 |
| Storage Expansion | Additional NVMe (8TB) | PCIe/PCIe x4 |
| Network Expansion | 10GbE, WiFi 6E | PCIe/PCIe x4 |
| I/O Expansion | USB hubs, GPIO banks | PCIe/PCIe x4 |

**Key Principle:** Each unit is self-contained. Stacking adds capacity, not dependency.

---

## Use Cases

### 1. Single Unit (Personal Athena)

**Setup:** 1x Athena NUC (ARM or x86)

**Use Case:**
- Personal AI assistant
- Local knowledge graph
- Game testing (with GPU option)
- Small-scale simulations

**Workload:**
- Text/audio/video ingestion
- Local inference (no cloud)
- Privacy-focused deployment

### 2. Dual Stack (Enhanced Athena)

**Setup:** 2x Athena NUC (CPU + NPU expansion)

**Use Case:**
- Enterprise deployment
- Larger knowledge graphs
- Parallel inference
- Redundancy/failover

**Workload:**
- 10x more nodes in graph
- 2x inference throughput
- High-availability operation

### 3. Quad Rack (Distributed Athena)

**Setup:** 4x Athena NUC (full rack)

**Use Case:**
- Research cluster
- Large-scale simulations
- Distributed manufacturing control
- Multi-site deployment

**Workload:**
- 100x more nodes
- 4x inference throughput
- Geographic distribution

### 4. OSE Integration

**Setup:** 1x Athena NUC + OSE machines

**Use Case:**
- Manufacturing control
- Machine monitoring
- Quality verification
- Autonomous fabrication

**Connection:**
- GPIO → OSE machine control
- USB → Machine interfaces
- Network → Remote machines

---

## Bill of Materials (BOM)

### Base Configuration (ARM)

| Component | Model | Qty | Unit Cost | Total |
|-----------|-------|-----|-----------|-------|
| CPU/SoC | BCM2712 | 1 | $80 | $80 |
| CPU RAM | 8GB LPDDR4X | 1 | $25 | $25 |
| NPU | Hailo-8L | 1 | $150 | $150 |
| NPU RAM | 8GB LPDDR5 | 1 | $30 | $30 |
| System SSD | 1TB NVMe | 1 | $100 | $100 |
| Graph SSD | 2TB NVMe | 1 | $200 | $200 |
| Bridge (FPGA) | Lattice ECP5 | 1 | $40 | $40 |
| Power System | USB-C PD | 1 | $30 | $30 |
| PCB | 8-layer | 1 | $60 | $60 |
| Enclosure | CNC Aluminum | 1 | $80 | $80 |
| Connectors | Various | 1 | $25 | $25 |
| **Total BOM** | | | | **$820** |

**Target Price:** $1,299 (37% margin)

### Performance Configuration (x86)

| Component | Model | Qty | Unit Cost | Total |
|-----------|-------|-----|-----------|-------|
| CPU | Ryzen 5 7600U | 1 | $200 | $200 |
| CPU RAM | 16GB DDR5 | 1 | $60 | $60 |
| NPU | Hailo-8 | 1 | $300 | $300 |
| NPU RAM | 8GB LPDDR5 | 1 | $30 | $30 |
| System SSD | 2TB NVMe | 1 | $200 | $200 |
| Graph SSD | 4TB NVMe | 1 | $400 | $400 |
| Bridge (FPGA) | Lattice ECP5 | 1 | $40 | $40 |
| Power System | USB-C PD | 1 | $30 | $30 |
| PCB | 10-layer | 1 | $80 | $80 |
| Enclosure | CNC Aluminum | 1 | $100 | $100 |
| Connectors | Various | 1 | $25 | $25 |
| **Total BOM** | | | | **$1,465** |

**Target Price:** $2,299 (36% margin)

---

## Software Stack

### CPU/GPU Domain

**Operating System:**
- Ubuntu 24.04 LTS (x86) or Raspberry Pi OS (ARM)
- Windows (optional, for game testing)

**Athena Server:**
- Python runtime
- API gateway
- Network management
- rPI coordination

**Applications:**
- Game engines (Unity, Unreal)
- Simulation software
- User interface
- Monitoring dashboards

### NPU Domain

**Firmware:**
- Bare-metal (Hailo-8L)
- Athena NPU runtime
- Graph storage manager
- Subgraph loader

**Models:**
- Encoder weights
- Projection weights
- VAE weights
- GNN weights

**Knowledge Graph:**
- Node embeddings
- Edge storage
- HNSW index
- FTS5 index

### Bridge Protocol

**Athena Inter-Domain Protocol (AIDP):**

```python
# CPU Side
class AthenaBridge:
    def send_query(self, query: bytes) -> bytes:
        """Send query to NPU, receive result."""
        pass
    
    def send_model_update(self, weights: bytes) -> bool:
        """Send model weights to NPU."""
        pass
    
    def get_status(self) -> NPUStatus:
        """Get NPU status."""
        pass
    
    def list_models(self) -> List[str]:
        """List available models on NPU."""
        pass
```

```c
// NPU Side (firmware)
void handle_bridge_message(Message* msg) {
    switch (msg->type) {
        case QUERY:
            process_query(msg->payload, msg->length);
            break;
        case UPDATE:
            load_model_weights(msg->payload, msg->length);
            break;
        case STATUS:
            send_status_response();
            break;
        case LIST_MODELS:
            send_model_list();
            break;
    }
}
```

---

## Development Roadmap

### Phase 1: Prototype (Q4 2026)

**Tasks:**
- [ ] Finalize NUC form factor design
- [ ] Design PCB (KiCad)
- [ ] Source components
- [ ] Assemble prototype
- [ ] Basic bridge testing

**Deliverables:**
- Working prototype
- Documentation
- Initial benchmarks

### Phase 2: Refinement (Q1 2027)

**Tasks:**
- [ ] Thermal optimization (passive cooling)
- [ ] Bridge protocol refinement
- [ ] Stacking validation
- [ ] Driver development
- [ ] Performance tuning

**Deliverables:**
- Optimized design
- Complete documentation
- Stacking demo

### Phase 3: Production (Q2 2027)

**Tasks:**
- [ ] Manufacturing partners
- [ ] Test fixtures
- [ ] Quality control
- [ ] Initial production run
- [ ] Beta testing

**Deliverables:**
- Production-ready design
- Manufacturing docs
- Beta units

### Phase 4: Launch (Q3 2027)

**Tasks:**
- [ ] Marketing materials
- [ ] Sales channels
- [ ] Support infrastructure
- [ ] Community launch

**Deliverables:**
- Commercial product
- Support system
- Developer community

---

## Open Source

**Hardware:** CERN-OHL-P (open hardware)
**Firmware:** MIT License
**Software:** MIT License

**Files to Release:**
- KiCad PCB files
- Schematics
- BOM
- Firmware source
- Bridge FPGA Verilog
- Driver source

---

*Document prepared for Athena NUC development.*
*Last updated: 2026-08-26*