The Python code doesn't show the value of nRow or from where it is obtained.

And your words:

reading file 05398.bin with size 90942 with nCols 3954 and nRows 23 nCols stays static for all of the files, but nRows changes for each file.

contradicts the evidence of the Python code which has nRow preset (somewhere) and calculates the value of nCols.

But, taking you at your word, and assuming the files consist of N rows of 3954 columns, you could do this:

use constant NCOLS => 3954;

my $file = $matrixPath.'/'.$info{$transcriptName}{'PATH'};
my $fileSize = -s $file;
die 'Bad filesize' unless $filesize % NCOLS == 0;

open IN, $file or die "Can't open inputfile $file $!";
binmode(IN);
## Note: You're reading the whole file in a single read NO NEED FOR A 
+LOOP.
sysread(IN, my $buffer, $fileSize) or die.
close IN;

## The first (rightmost) unpack splits the buffer into NCOLS length se
+ctions of bytes.
## The second unpack breaks each of those into its individual (uchar) 
+integers 
## and puts them in an anonymous array. 
## the map assigns al the anonymous arrays to @matrix.

# Update: template corrected, see posts below.
my @matrix = map[ unpack 'C*', $_ ], unpack '(a' . NCOLS . ')*', $buff
+er;
...
[download]