PCA9655E breakout

04 February 2018
While I was doing the reflow for the PCA9671 expander breakout I also used the same PCB board to mount an ON Semiconductor PCA9655E (Farnell 2464421) since aside from having an interrupt pin rather than a reset, the pin-outs are virtually identical. As it turned out this I/O expander is functionally identical to the TCA9535, while having the higher total current draw tolerance and address range of the PCA9671. As a result this article will highlight the differences.

The PCA9655E is register-based where all I²C transactions, including pin reads & writes, use an internal 8-byte address. The 16 input/output pins are divided into two 8-bit ports, which operate independently. Each pin can sink 50mA with a combined ground current limit of 600mA, although there is a recommended limit of 25mA per pin.

Usage

I²C communication of the expander makes use of the same register addresses as the TCA9535, so the sub-sections below will only cover the difference. One nice feature is the ability to use a multi-byte transaction to read/write more than one register at once — I did not check at the time whether the TCA9535 also allows this — which is particularly useful as it allows one transaction to deal with all 16 pins over the two ports.

Address pins

Like the PCA9671, the PCA9655E expander uses the I²C SDA & SCL lines as input for the device address selection pins in addition to V_cc and Ground. I am not exactly sure how it works, but my suspicion it is an elimination process based on comparing when the pins change state, assuming they change at all. There is quite likley coordination between manufacturers as the address table in the data-sheet has the same mapping as that in the PCA9671 data-sheet, and is replicated below:

Address pins			I²C address
A2	A1	A0	Write	Read
Gnd	Scl	Gnd	0x20	0x21
Gnd	Scl	V_cc	0x22	0x23
Gnd	Sda	Gnd	0x24	0x25
Gnd	Sda	V_cc	0x26	0x27
V_cc	Scl	Gnd	0x28	0x29
V_cc	Scl	V_cc	0x2a	0x2b
V_cc	Sda	Gnd	0x2c	0x2d
V_cc	Sda	V_cc	0x2e	0x2f
Gnd	Scl	Gnd	0x30	0x31
Gnd	Scl	V_cc	0x32	0x33
Gnd	Sda	Gnd	0x34	0x35
Gnd	Sda	V_cc	0x36	0x37
V_cc	Scl	Gnd	0x38	0x39
V_cc	Scl	V_cc	0x3a	0x3b
V_cc	Sda	Gnd	0x3c	0x3d
V_cc	Sda	V_cc	0x3e	0x3f
Gnd	Gnd	Gnd	0x40	0x41
Gnd	Gnd	V_cc	0x42	0x43
Gnd	V_cc	Gnd	0x44	0x45
Gnd	V_cc	V_cc	0x46	0x47
V_cc	Gnd	Gnd	0x48	0x49
V_cc	Gnd	V_cc	0x4a	0x4b
V_cc	V_cc	Gnd	0x4c	0x4d
V_cc	V_cc	V_cc	0x4e	0x4f
Gnd	Gnd	Scl	0x50	0x51
Gnd	Gnd	Sda	0x52	0x53
Gnd	V_cc	Scl	0x54	0x55
Gnd	V_cc	Sda	0x56	0x57
V_cc	Gnd	Scl	0x58	0x59
V_cc	Gnd	Sda	0x5a	0x5b
V_cc	V_cc	Scl	0x5c	0x5d
V_cc	V_cc	Sda	0x5e	0x5f
Scl	Scl	Gnd	0xa0	0xa1
Scl	Scl	V_cc	0xa2	0xa3
Scl	Sda	Gnd	0xa4	0xa5
Scl	Sda	V_cc	0xa6	0xa7
Sda	Scl	Gnd	0xa8	0xa9
Sda	Scl	V_cc	0xaa	0xab
Sda	Sda	Gnd	0xac	0xad
Sda	Sda	V_cc	0xae	0xaf
Scl	Scl	Scl	0xb0	0xb1
Scl	Scl	Sda	0xb2	0xb3
Scl	Sda	Scl	0xb4	0xb5
Scl	Sda	Sda	0xb6	0xb7
Sda	Scl	Scl	0xb8	0xb9
Sda	Scl	Sda	0xba	0xbb
Sda	Sda	Scl	0xbc	0xbd
Sda	Sda	Sda	0xbe	0xbf
Scl	Gnd	Gnd	0xc0	0xc1
Scl	Gnd	V_cc	0xc2	0xc3
Scl	V_cc	Gnd	0xc4	0xc5
Scl	V_cc	V_cc	0xc6	0xc7
Sda	Gnd	Gnd	0xc8	0xc9
Sda	Gnd	V_cc	0xca	0xcb
Sda	V_cc	Gnd	0xcc	0xcd
Sda	V_cc	V_cc	0xce	0xcf
Scl	Gnd	Scl	0xe0	0xe1
Scl	Gnd	Sda	0xe2	0xe3
Scl	V_cc	Scl	0xe4	0xe5
Scl	V_cc	Sda	0xe6	0xe7
Sda	Gnd	Scl	0xe8	0xe9
Sda	Gnd	Sda	0xea	0xeb
Sda	V_cc	Scl	0xec	0xed
Sda	V_cc	Sda	0xee	0xef

Addresses ranges 0x00-0x1e, 0xd0-0xde, and 0xf0-0xfe are not used. Some of these correspond to I²C reserved addresses which ought not be used anyway.

Interrupts

Due to a PCB design error I was not able to verify the expected interrupt behaviour of the PCA9671, but since no such error exist with this breakout board, I did a quick test of the interrupt pin using the setup shown below. The interrupt pin is active-low open-drain, and it resets either when the port with the changed pin is read or the pins return to the pre-interrupt state.

I am not exactly sure how it resolves the reset when pins on both have changed but only one port is read, so I would be inclined to read both ports with a two-byte read:

./ttyTxRx.py /dev/ttyACM0 2 55 41 00 2

Remarks

While it was a nice discovery that the command set was identical to an I/O expander I had tried out in the past, I find the command-orientated ones that use register addresses to be a bit of a pain. The Microchip MCP23017 seemed to include 1 or 2 fringe options purely because there was the spare capacity, but the others I have tried are not quite so pathological. Personally I prefer the zero-configuration nature of the 8-bit PCF8574 and 16-bit PCA9671, the latter being physically identical to the expander this article details.