DLPC3420: DLP Controlled does not initialize, HOST_IRQ remains tri-stated

Hello,

We are developing a custom DMD module with the DLPC3420 Controller, DLPA2005 PMIC and DLP160AP DMD.

It appears that the DLP Controller does not initialize on our prototype, as the HOST_IRQ remains high (when pulled up) and the device is not responding to I2C communication.
The DMD Is currently disconnected but we have observed the same behavior with it connected. 

We have confirmed:

• All power supplies are stable at correct voltages as provided by the PMIC.
• The control signals RESETZ, PARKZ and PROJ_ON are supplied in the correct timing by the PMIC.
• The device draws about 10-20 mA of current when powered and PROJ_ON is applied. 
• The PLL is running stable at the correct frequency (24 MHz) when powered
• The device is communicating with the flash memory (we are using Macronix MX25L1606E) after being powered up. The controller first reads the initial 16 bytes at low frequency, the switches to high frequency and continues reading the flash contents. We have verified that the flash responds and supplies data which matches the firmware image file to the controller.
• JTAGRSTZ is pulled low via a 1k resistor to ground.
• HWTEST_EN is tied directly to ground
• The I2C lines are pulled high at all times (during our testing, except when testing the I2C communication)
• Other signals (Parallel/DSI interface, JTAG, GPIO except PROJ_ON and RC_CHARGE) are left floating or unconnected
• The controller is not communicating with the PMIC

During testing, we have disconnected the pullup resistor from the HOST_IRQ signal to observe the driving behavior. According to the datasheet, the controller should drive HOST_IRQ high at the start of initialization, but this does not happen. The signal remains tri-stated at all times. As far as I understand, this signal should be driven high when the initialization starts and we know that it starts, because the device is reading the contents of flash, but once again, it remains tri-stated instead. 

We have a second matching prototype which behaves in the exactly same way.

Any insights are greatly appreciated