Skip to content

Relay Instead of Router

"Locality scales. Global connectivity creates chaos."

In a world where "more connections = better", we chose the path of constraints. And precisely this gives Decima-8 an advantage where others choke on complexity.

Essence

Approach Architecture Consequence
Router (Loihi, traditional) Neuron connects to any other Exponential complexity growth, jitter, collisions
Relay (Decima-8) Tile connects only to 8 neighbors (N/E/S/W/NE/SE/SW/NW) Linear growth, 0 jitter, determinism

Relay is not a "limitation". It is an architectural filter that eliminates chaos at the topology level.

Why Routing Doesn't Scale

Problem What Happens Cost
Packet collisions Two signals want to pass through one node Arbitration, delays, non-determinism
Routing "Where to send this packet?" Computational costs, logic
Buffering Packet queue in router Memory, delays, jitter
Verification "Why did the packet arrive with delay?" Impossible to trace all paths

In biology this is acceptable (brain is asynchronous, jitter is not critical). In engineering ā€” no (HFT, robotics, medicine require predictability).

In Code

// Relay: each tile knows only 8 neighbors
typedef struct {
    uint8_t activation;
    uint8_t relay_mask;  // Bit mask: N=1, E=2, S=4, W=8
} Tile;

// Signal transmission ā€” deterministic shift
void relay(Tile* tile, Tile* neighbors) {
    if (tile->relay_mask & NORTH) neighbors[0].input += tile->output;
    if (tile->relay_mask & EAST)  neighbors[1].input += tile->output;

Contrast with routing:

// Router: where to send?
Route path = find_route(source, destination);  // Path finding
if (router.busy) { enqueue(packet); }          // Queue
if (collision) { resolve_conflict(); }         // Arbitration
send(packet, path);                            // Send

// Result: delay is unpredictable (jitter)

Philosophically

"Freedom of choice ā‰  freedom of connections."

In biology, a neuron can connect to any other ā€” this is plasticity. In engineering, a tile connects only to neighbors ā€” this is predictability.

Principle Biology Engineering
Connectivity Global (through axons) Local (8 neighbors)
Goal Survival in chaos Predictability in order
Scaling Evolution filters out errors Design excludes errors

Relay is trust in the local. Each tile knows only its neighbor. But the entire network knows the goal.

Practically

Advantage How it helps in B2B
Scaling Want more tiles? Just add them. Architecture doesn't change.
Energy No router = no routing costs (up to 40% savings)
Verification Every path is predictable ā†’ can prove correctness
Debugging Signal passed through Nā†’Eā†’Sā†’W. Tracing is unambiguous.
ASIC Simple topology = fewer metal layers = cheaper chip

Scaling: Linear vs Exponential

Number of Tiles Router (complexity) Relay (complexity)
256 O(n log n) O(n)
4,096 O(nĀ²) O(n)
1,048,576 (1M) Impractical Linear

This is not "simplification", but an architectural advantage that allows scaling to 1M+ tiles without complexity explosion.

Real-World Analogy

System Approach Result
Internet Packet routing, routers Flexible, but jitter, losses
Assembly Line Part moves from machine to machine Predictable, deterministic
Decima-8 Relay between tiles Conveyor belt for patterns

For communication ā€” you need the internet. For manufacturing ā€” you need an assembly line.

We propose building personality manufacturing.

Read Further