diff --git a/litex/soc/cores/i2s.py b/litex/soc/cores/i2s.py
index 96011ad3e..f0554d5f2 100644
--- a/litex/soc/cores/i2s.py
+++ b/litex/soc/cores/i2s.py
@@ -11,9 +11,7 @@ from litex.soc.integration.doc import AutoDoc, ModuleDoc
 
 class S7I2SSlave(Module, AutoCSR, AutoDoc):
     def __init__(self, pads, fifo_depth=256):
-        self.intro = ModuleDoc("""
-        Intro
-        *******
+        self.intro = ModuleDoc("""Intro
 
         I2S slave creates a slave audio interface instance. Tx and Rx interfaces are inferred based
         upon the presence or absence of the respective pins in the "pads" argument.
@@ -24,7 +22,7 @@ class S7I2SSlave(Module, AutoCSR, AutoDoc):
         to a CODEC without a programmable bit offset!
 
         System Interface
-        =================
+        ----------------
 
         Audio interchange is done with the system using 16-bit stereo samples, with the right channel
         mapped to the least significant word of a 32-bit word. Thus each 32-bit word is a single
@@ -56,12 +54,12 @@ class S7I2SSlave(Module, AutoCSR, AutoDoc):
         - enable the Tx FIFO to run
         - poll or wait for interrupt; upon interrupt, write `fifo_depth` words. Repeat.
         - to close stream, mute the DAC and stop the request clock. Ideally, this can be completed
-        before the FIFO is emptied, so there is no jarring pop or truncation of data
+          before the FIFO is emptied, so there is no jarring pop or truncation of data
         - stop FIFO running. Next initiation should reset the FIFO to ensure leftover previous data
-        in FIFO is cleared.
+          in FIFO is cleared.
 
         CODEC Interface
-        ================
+        ---------------
 
         The interface assumes we have a sysclk domain running around 100MHz, and that our typical max
         audio rate is 44.1kHz * 24bits * 2channels = 2.1168MHz audio clock. Thus, the architecture
@@ -87,12 +85,11 @@ class S7I2SSlave(Module, AutoCSR, AutoDoc):
         - Data is updated on the falling edge
         - Data is sampled on the rising edge
         - Words are MSB-to-LSB, left-justified (**NOTE: this is a deviation from strict I2S, which
-        offsets by 1 from the left**)
+          offsets by 1 from the left**)
         - Sync is an input (FPGA is slave, codec is master): low => left channel, high => right channel
         - Sync can be longer than the wordlen, extra bits are just ignored
         - Tx is data to the codec (SDI pin on LM49352)
         - Rx is data from the codec (SDO pin on LM49352)
-
         """)
 
         # One cache line is 8 32-bit words, need to always have enough space for one line or else
diff --git a/litex/soc/cores/spi_opi.py b/litex/soc/cores/spi_opi.py
index 1e43882b9..45f477677 100644
--- a/litex/soc/cores/spi_opi.py
+++ b/litex/soc/cores/spi_opi.py
@@ -18,9 +18,7 @@ class S7SPIOPI(Module, AutoCSR, AutoDoc):
         sim            = False,
         spiread        = False,
         prefetch_lines = 1):
-        self.intro = ModuleDoc("""
-        Intro
-        ********
+        self.intro = ModuleDoc("""Intro
 
         SpiOpi implements a dual-mode SPI or OPI interface. OPI is an octal (8-bit) wide variant of
         SPI, which is unique to Macronix parts. It is concurrently interoperable with SPI. The chip
@@ -300,14 +298,13 @@ class S7SPIOPI(Module, AutoCSR, AutoDoc):
             ),
         ]
 
-        self.architecture = ModuleDoc("""
-        Architecture
-        **************
+        self.architecture = ModuleDoc("""Architecture
 
         The machine is split into two separate pieces, one to handle SPI, and one to handle OPI.
 
         SPI
-        =====
+        -----
+
         The SPI machine architecture is split into two levels: MAC and PHY.
 
         The MAC layer is responsible for:
@@ -333,7 +330,8 @@ class S7SPIOPI(Module, AutoCSR, AutoDoc):
         Thus holding "req" high can allow the PHY to back-to-back issue cycles without pause.
 
         OPI
-        =====
+        -----
+
         The OPI machine is split into three parts: a command controller, a Tx PHY, and an Rx PHY.
 
         The Tx PHY is configured with a "dummy cycle" count register, as there is a variable length
@@ -389,6 +387,7 @@ class S7SPIOPI(Module, AutoCSR, AutoDoc):
              - pre-fetch is aborted because bus_adr and next read address don't match and FIFO is reset
 
            RxPHY:
+
            - while CTI==2, assemble data into 32-bit words as soon as EMPTY is deasserted,
              present a bus_ack, and increment the next read address pointer
            - when CTI==7, ack the data, and wait until the next bus cycle with CTI==2 to resume