> ## Documentation Index
> Fetch the complete documentation index at: https://docs.thrindex.com/llms.txt
> Use this file to discover all available pages before exploring further.

# E0401 — LifNotSupported

> A LIF layer was found in an artifact compiled for akida-akd1500. AKD1500 implements bounded ReLU, not leaky integrate-and-fire.

<Warning>E0401: LIF layer cannot be mapped to akida-akd1500</Warning>

## What happened

A `.thx` artifact was validated against the `akida-akd1500` backend and one of its layers has `type: "lif"`. The AKD1500 hardware does not implement leaky integrate-and-fire (LIF) dynamics and cannot execute this layer.

## Why

AKD1500 implements **Akida 1.0**. Its activation function is a bounded ReLU applied per-inference-call to an integer dot product. It has:

* **No membrane potential** — there is no state variable that accumulates across time.
* **No exponential leak** — the `alpha = exp(-dt / tau_mem)` term in the LIF equation has no hardware equivalent.
* **No spike-triggered reset** — the subtract or zero reset modes have no hardware equivalent.

LIF neurons and bounded ReLU are different computational models. There is no lossless conversion between them.

## How to fix

**Option A — Use the simulator.** If you need SNN inference with LIF dynamics, compile for `target="sim"`:

```python theme={null}
thx.compile(model, "model.thx", target="sim")
```

The simulator implements LIF fully, including membrane potential, synaptic dynamics, and both reset modes.

**Option B — Redesign for AKD1500.** AKD1500 runs feedforward integer networks (Dense → ReLU). If you want hardware acceleration and do not need SNN temporal dynamics, redesign the model without LIF layers. Refer to the [BrainChip documentation](https://doc.brainchipinc.com/) for the supported layer types.

There is no automatic model conversion. A model that uses LIF neurons depends on temporal state that does not exist in the AKD1500 execution model.

## Example

```
E0401: layer[1] type="lif" cannot be mapped to akida-akd1500.
Observed: .thx layer 1 has type "lif" with threshold=1.0, alpha=0.905, reset="subtract".
Why: AKD1500 implements Akida 1.0. Akida 1.0's activation is bounded ReLU applied
     per-inference-call to the integer dot product. It has no membrane potential, no
     exponential leak (alpha term), and no spike-triggered reset.
What to do: (a) Use the 'sim' backend for SNN simulation.
            (b) If you want AKD1500 inference, re-design the model without LIF layers.
Docs: https://docs.thrindex.com/errors/E0401
```